POPULARITY
87 episodes with annotations
15 episodes with annotations
6 episodes with annotations
18 episodes with annotations
12 episodes with annotations
5 episodes with annotations
5 episodes with annotations
3 episodes with annotations
3 episodes with annotations
There's a reason most people never become known for their thought leadership. It's not because they lack ideas… But because they haven't done the work to build their credibility or subject matter expertise. Today's guest is Simone Heng, a human connection specialist and award-winning author. Simone's mission is to inspire deeper connection in a world that's increasingly disconnected. Her book Let's Talk About Loneliness has received six international accolades. She's also spoken to thousands (ex: Harvard, Google, Meta, the United Nations, and more), and her work has been featured by CNN, Al Jazeera, Forbes, Harvard Business Review, and BBC Radio. Simone built her career in thought leadership not through shortcuts, but through years of intentional work - Educating herself, building a credible brand, and publishing ideas that are not only polished, but profound. If you've ever wondered what it looks like to play the long game in a world that rewards speed… This is the conversation. In this episode, we cover: (00:00) Introduction (01:57) “I've never seen an Asian woman speak like that” (03:13) Developing a subject matter expertise (07:17) Choosing your topic of expertise (09:20) Being told to hide parts of you (10:41) The behind the scenes work (12:36) Your story isn't enough (16:09) Do you need a PhD to be a thought leader? (22:24) Reviewing Simone's past content (26:52) Become a speaker or author (30:49) What's next for Simone (35:43) Actionable Takeaways Connect with Simone Heng Instagram: https://www.instagram.com/simoneheng LinkedIn: https://www.linkedin.com/in/simone-heng-speaker Website: https://simoneheng.com Thought Leaders Now: https://www.thoughtleadersnow.com Connect with Cheryl Lau Website: https://cheryllau.com YouTube: https://www.youtube.com/@cheryltheory Instagram: https://www.instagram.com/cheryltheory LinkedIn: https://www.linkedin.com/in/cheryllau ANNOTATIONS - The Newsletter The edits that got cut from the podcast. Every interview on EDIT HISTORY runs about 60 minutes. But less than 40 minutes makes it into the final cut. This newsletter is where the rest live. ANNOTATIONS is where I share the 33% I left behind — and the insights that came after we stopped recording. Subscribe at: https://cheryllau.com/email Contact Please email hello@cheryllau.com for business inquiries.
There's a moment when we're taught to quiet ourselves. Maybe someone said they couldn't understand your accent. Maybe you were told you were “too much.” Or maybe something happened that made you start playing small just to stay safe. And so your voice - your real voice - gets buried under layers of doubt, conditioning, and fear… But what happens when you decide you're done with that? Today's guest is Jam Gamble, an educator, award-winning media personality, and one of the fiercest advocates for vocal empowerment I've ever met. She's the CEO behind Slay The Mic, a program that helps speakers, content creators, and entrepreneurs transform how they use their voice. In this episode, we talk about reclaiming your voice after it's been silenced, why your story is more powerful than you think, and how giving yourself permission to be heard might just be the most pivotal thing you do. Because somewhere out there, someone is waiting for you to go first. In this episode, we cover: (00:00) Introduction (01:39) Who's Your Kanye? (Framework) (03:04) Who was Jam's Kanye? (05:30) 3 Types of Kanye (08:40) Acknowledging the elephant in the room (11:42) Resonating with your audience (14:47) Reclaiming your voice (23:56) Reviewing Jam's past content (31:39) Vocal roadblocks (32:41) Speaking with an accent (36:31) You ARE the example (41:38) Actionable Takeaways Connect with Jam Gamble Instagram: https://www.instagram.com/iamjamgamble Website: https://www.iamjamgamble.com Connect with Cheryl Lau Website: https://cheryllau.com YouTube: https://www.youtube.com/@cheryltheory Instagram: https://www.instagram.com/cheryltheory LinkedIn: https://www.linkedin.com/in/cheryllau ANNOTATIONS - The Newsletter The edits that got cut from the podcast. Every interview on EDIT HISTORY runs about 60 minutes. But less than 40 minutes makes it into the final cut. This newsletter is where the rest live. ANNOTATIONS is where I share the 33% I left behind — and the insights that came after we stopped recording. Subscribe at: https://cheryllau.com/email Contact Please email hello@cheryllau.com for business inquiries.
As creators, many of us start with a hunch, a curiosity, or a simple “let's see where this goes.” Today's guest, Dexter Zhuang, embodies this experimental mindset. He's the founder of Portfolio Path and a fractional product leader who's spent the last 12 years leading teams across the U.S., Southeast Asia, and Latin America at companies like Dropbox and Xendit. In 2023, Dexter launched Money Abroad—a newsletter exploring personal finance for expats—as an experiment. Over time, it evolved into Portfolio Path, an education platform that helps high-performers grow their portfolio careers and manage their money with intention. Today, Portfolio Path reaches over 7,000 subscribers. Along the way, Dexter has leaned into audience surveys, embraced feedback, and navigated the common insecurities that come with publishing your ideas publicly. In this episode, we unpack that journey—from testing an idea to building something that resonates deeply with a growing community. In this episode, we cover: (00:00) Introduction (02:02) Starting a newsletter as an experiment (06:11) Pivoting the newsletter (11:10) Reviewing Dexter's past content (16:26) Insecurities when building the newsletter (20:49) Content is a stepping stone for the audience (23:24) Conducting surveys with your audience (30:52) Underrated parts of building a newsletter (37:26) Actionable Takeaways Connect with Dexter Zhuang Newsletter: https://www.theportfoliopath.com/subscribe Website: https://www.dexterzhuang.com LinkedIn: https://www.linkedin.com/in/dexterzhuang Connect with Cheryl Lau Website: https://cheryllau.com YouTube: https://www.youtube.com/@cheryltheory Instagram: https://www.instagram.com/cheryltheory LinkedIn: https://www.linkedin.com/in/cheryllau ANNOTATIONS - The Newsletter The edits that got cut from the podcast. Every interview on EDIT HISTORY runs about 60 minutes. But less than 40 minutes makes it into the final cut. This newsletter is where the rest live. ANNOTATIONS is where I share the 33% I left behind — and the insights that came after we stopped recording. Subscribe at: https://cheryllau.com/email Contact Please email hello@cheryllau.com for business inquiries.
L'enrichissement vidéo sans complexité (et sans IA) ! Dans ce nouvel épisode de Rendez-vous en terre digitale, Clément et Olivier
Send us a textNew forms of life that were never meant to be. Drugs that open doorways to the other side. And a Juggalo who winds up at the mercy of a cult. This just scratches the surface of the darkness this story has to offer, Join us, and come down below...Annotations provided by Official Show Archivist Mary Anne Simpson.After this, I encourage you to listen toThicker Than Water, a new audio novel by yours truly. 11 hours of crime noir goodness, a savage tale of revenge, and family. I will be releasing the first couple of chapters right here, very soon. Its available for free on the patreon, but its also for sale! 10 bucks, no membership required. Music by:Serge Quadrado - documentary dark dunJIN – a blood red thing that writhesRoom of Wires – asylum sneakerFilmy Ghost – my mask is broken Please subscribe through Buzzsprout, Stitcher, Spotify, Podchaser, or iTunesFind me on social media on Instagram Facebook and Twitter, or email me direct at AScaryHomeCompanion@gmail.comSupport our PATREON page! And check out the Redbubble merch shop. Support the show
Join Fr. Jacob and Fr. Mike as they explore the power of the stories we tell—about others and ourselves. Inspired by Ignatius' Annotation 22 and Fr. Jacob's recent read, Viper's Tongue, this thought-provoking discussion dives into how our perceptions shape relationships, influence motives, and impact the narratives others create about us and our narratives about others.
Send us a textA black ops strike team prowls the darkest shadows of America. They are hunting Breach Sites, those soft spots where a door to hell has once been opened. They seek to close them all, permanently. The ghastly cultists have other ideas. Action and suspense abound, and god help the wicked.Annotations provided by Official Show Archivist Mary Anne Simpson.After this, I encourage you to listen toThicker Than Water, a new audio novel by yours truly. 11 hours of crime noir goodness, a savage tale of revenge, and family. I will be releasing the first couple of chapters right here, very soon. Its available for free on the patreon, but its also for sale! 10 bucks, no membership required. Please subscribe through Buzzsprout, Stitcher, Spotify, Podchaser, or iTunesFind me on social media on Instagram Facebook and Twitter, or email me direct at AScaryHomeCompanion@gmail.comSupport our PATREON page! And check out the Redbubble merch shop. Support the show
Mary Oliver, Frank O'Hara and Like Streams to the Ocean
In this thought-provoking episode, I sit down with Dr. Remi Kalir, the Associate Director of Faculty Development and Applied Research with Learning Innovation and Lifetime Education at Duke University, where he also serves as Associate Director of the Center for Applied Research and Design in Transformative Education. He has also completely revolutionized my thinking about annotation. As someone who was relatively ambivalent about annotations, Remi's perspective transformed me into a fan, believer, and enthusiastic practitioner. Our conversation challenges conventional wisdom about annotation, as Remi argues that we're all annotators, from the grandmother scribbling recipe modifications to fans dissecting Kendrick Lamar's lyrics on Genius. He also shares fascinating examples from his upcoming book "Re/Marks on Power" (MIT Press, 2025), including Harriet Tubman's previously unexamined annotations in pension files, protest markings on Confederate monuments, and how the US-Mexico border itself represents a form of annotation—a line drawn imprecisely on a map as an exercise of power.Key Concepts from the Episode:Annotation as a Social PracticeAnnotation is more than a reflection of individual comprehensionAnnotations have a "social life" that extends beyond the text and timeAnnotation is dialogic rather than an isolated literacy actAnnotation as a Tool for CritiqueAnnotation serves as a tool for critique and challenging authorityAnnotation can circulate counter-narratives and resist dominant ideologiesE.g. Harriet Tubman's use of annotations on pension documentsAnnotation as an Embodied PracticeAnnotations can be embodied and geographic Protests and interventions on monuments represent forms of annotationDigital annotation practices are all over spaces like TikTok, Genius, etc.Particularly compelling is our discussion of annotation's unique affordances: its proximity to the original text, its capacity for "rough draft thinking," and its ability to make our responses visible to others across time and space. Remi invites us to see annotation not as an isolated comprehension check but as a dialogic practice with profound implications for critical literacy, social justice, and civic engagement. For educators struggling to make annotation meaningful beyond compliance, this episode offers both theoretical insights and practical inspiration to transform this everyday practice into something that can, as Remi says, "live, speak, and inspire."Re/Marks on Power (Newsletter)Re/Marks on Power (Book)Join me and socially annotate the transcription!Support the show
Send us a textWelcome to the 21st edition of DigiPath Digest! In this episode, together with Dr. Aleksandra Zuraw you will review the latest digital pathology abstracts and gain insights into emerging trends in the field. Discover the promising results of the PSMA PET study for prostate cancer imaging, explore the collaborative open-source platform HistioColAI for enhancing histology image annotation, and learn about AI's role in improving breast cancer detection. Dive into topics such as the role of AI in renal histology classification, the innovative TrueCam framework for trustworthy AI in pathology, and the latest advancements in digital tools like QuPath for nephropathology. Stay tuned to elevate your digital pathology game with cutting-edge research and practical applications.00:00 Introduction to DigiPath Digest #2101:22 PSMA PET in Prostate Cancer06:49 HistoColAI: Collaborative Digital Histology12:34 AI in Mammogram Analysis17:21 Blood-Brain Barrier Organoids for Drug Testing22:02 Trustworthy AI in Lung Cancer Diagnosis30:09 QuPath for Nephropathology35:30 AI Predicts Endocrine Response in Breast Cancer40:04 Comprehensive Classification of Renal Histologic Types45:02 Conclusion and Viewer EngagementLinks and Resources:Subscribe to Digital Pathology Podcast on YouTubeFree E-book "Pathology 101"YouTube (unedited) version of this episodeTry Perplexity with my referral linkMy new page built with PerplexityHistoColAI Github PagePublications Discussed Today:
We built video annotations with tldraw! It's a new feature we're launching next week, and we're really excited for all you Jamming to try it. So today, Jam engineers: Max, Aidan, and Rui get into the technical details of implementing the tldraw library - so you can draw stuff while recording your screen.Excited to show you what we built!(00:37) Why implementing annotations was so different than the blur tool(02:55) How Max discovered we already had a tldraw license(04:25) Why we love tldraw: React-SVG dual architecture & more details(08:55) Demo of video annotations & why it's different than Jam's screenshot feature(11:29) Why we ultimately decided to use tldraw for video too (it looks so nice!)(12:52) Our biggest takeaway for building w/ 3rd party librariesSubscribe to Building Jam on YouTube, Spotify, and Apple Podcasts. New episodes drop every Friday at 10AM ET. See you there!
CraftLit - Serialized Classic Literature for Busy Book Lovers
FOR ACTUAL TEXT of Mary Wollstonecraft's *A Vindication of the Rights of Woman* USE CLOSED CAPTIONS—Links to discussed topics can be found below the timecodes. Please add QUESTIONS and CLARIFICATIONS in the comments. FULL SERIES: 00:00 Intro notes 01:56 Understanding MW's Annotations 05:02 Note 1 on Islam - PBS LINK: 05:44 Sura Ghafir 40:40 06:17 Note on Islam 2 07:03 Define: Providence 08:00 Notes on Genesis and Creation 10:07 Definition: Sensual 11:01 Virtue and Moratily in MWs View 12:22 Deifine: Positive 14:23 CHAPTER TWO TEXT: A Vindication on the Rights of Woman 14:30 The Prevailing Opinion of a Sexual Character Discussed 16:07 Women's Education and Virtue 18:09 Critique of Rousseau and Other Authors 28:16 The Superficial Knowledge of Women and Soldiers 31:51 The Impact of Standing Armies 34:57 The Tyranny of Sensualists 35:32 Rousseau's Unnatural Sophia 42:28 Post-chapter Footnotes *Links for you* Not Discussed Today, But Useful Now & In Future Episodes , written within a year after her death from Placental Sepsis after giving birth to her second daughter Mary Wollstonecraft Godwin (Shelley) Mary Wollstonecraft's death: Link below Full-Text Links Full-text of Vindication: Full-text of Paradise Lost: Full-text of Èmile: More links and info on MW's death can be found at the end of this post. Not Wollstonecraft, but also good to know... Bot Army—Irksome Humans May Not Be Human My response: Ages ago, there was a Twitter bot that you could forward a tweet to and get a reading of a % chance whether or not the tweet came from a bot. I used it all the time—and calmed down A LOT. Then it disappeared. Does anyone else remember using something like that? CraftLit's Socials Find everything here: Join the newsletter: Podcast site: Facebook: Facebook group: Pinterest: TikTok podcast: Spooky Narration: Email: heather@craftlit.com Call and share your thoughts! 1-206-350-1642 SUPPORT THE SHOW! CraftLit App Premium feed (only one tier available) PATREON: (all tiers, below) ——Walter Harright - $5/mo for the same audio as on App ——Jane Eyre - $10/mo for even-month Book Parties ——Mina Harker - $15/mo for odd-month Watch Parties All tiers and benefits are also available as YouTube Channel Memberships Ko-Fi NEW at CraftLit.com — Premium SITE Membership (identical to Patreon except more of your support goes to the CraftLit Team) If you want to join us for a particular Book or Watch Party but you don't want to subscribe, please use or CraftLit @ Venmo and include what you want to attend in the message field. Please give us at least 24 hours to get your message and add you to the attendee list. Download the FREE CraftLit App for iOS or Android (you can call or email feedback straight from within the app) Call 1-206-350-1642 __________ MW's Death Trigger Warning: Women's Healthcare—Placental sepsis Placental sepsis led to the death of Mary Wollstonecraft in 1797 after she gave birth to her daughter Mary Godwin. It is now more commonly known as puerperal sepsis or postpartum sepsis. This condition is an infection that occurs after childbirth. In the past, it was a major cause of maternal deaths related to childbirth, especially before modern hygiene practices and antibiotics became available. Global Situation Today: - Maternal sepsis remains a serious issue and is still a significant cause of maternal deaths around the world. - The World Health Organization (WHO) estimates that maternal sepsis accounts for about 10% of all maternal deaths globally. It tends to be more common in areas where many births happen at home, there are not enough skilled healthcare workers, and healthcare systems are weak. Historical Background: During Wollstonecraft's time, doctors often worked in unhygienic environments and did not yet understand germs. Consequently, infections after childbirth were sadly common and often turned deadly. - Peer Reviewed Journal Articles on Placental Sepsis: Cambridge: ; AIMDR: ; Incidences of: -Trigger Warning: Details on MW's death:
Page Annotation for Google iOS & More Digital Marketing News | Marketing O'Clock Episode 359 This week on Marketing O'Clock, Google turns the wrong page toward Page Annotation for the iOS app. The DOJ is doubling down on selling Chrome after their antitrust win against Google. Plus, Google launches Customer Match lists in GA4 helping marketers find their perfect match with audiences. Visit us at - https://marketingoclock.com/
Pascal Praud revient pendant deux heures, sans concession, sur tous les sujets qui font l'actualité. Vous voulez réagir ? Appelez-le 01.80.20.39.21 (numéro non surtaxé) ou rendez-vous sur les réseaux sociaux d'Europe 1 pour livrer votre opinion et débattre sur les grandes thématiques développées dans l'émission du jour.
Today, I'm talking about something I don't do well in my reading life, but I want to: adding annotation and marginalia to the books I read. I'm quite comfortable with my sticky note and book dart practices, but something is calling me to explore more methods to leave tracks of my thinking in the books I read. Come listen as I explore the how and why behind annotation, marginalia and what I like to call ‘bookish notes to self'. We'll discuss why it matters and I'll offer 5 ways you could bring the practice to your own reading life…which means you're more able to make changes in your actual life, too. What do YOU think about annotation, marginalia and bookish notes to self? I'd love to hear more about your practices in the comments so I can give them a try, too! You'll find the show notes for the episode with links to all of the books and resources mentioned right here: https://www.alitlife.com/2024/10/22/annotation-marginalia-bookish-notes-to-self/ Love this podcast and want more? Consider this your invitation to join my Get Lit(erate) Patreon community! Each month, we take a deep dive into one bookish theme and work to bring it to life in our own lives. You'll get bonus episodes, book calendars, live book club and notebook sessions, special events and much more. Learn more at www.getliterate.co. Get your own Get Lit(erate). notebook to take notes on the books you want to read and notebook ideas you want to try: https://amzn.to/44wELKN If you'd like to support the podcast, consider purchasing some Get Lit(erate). merchandise from my Zazzle store: https://www.zazzle.com/store/alitlife All earnings are funneled right back into the podcast expenses and maintenance fees. Thanks for your support! Follow Stephanie: Website: http://www.alitlife.com/ Facebook: http://www.facebook.com/AffinitoLit Twitter: http://www.twitter.com/AffinitoLit Instagram: http://www.instagram.com/AffinitoLit
Welcome to the Sterile Technique Podcast! It's the podcast about Surgical Technology. Whether you are a CST or CSFA, this podcast helps you earn CE credits and improve your surgery skills in the OR. This episode discusses the cover article of the October 2024 issue of The Surgical Technologist, which is the official journal of the Association of Surgical Technologists (AST). The article is titled, "Palatal Hybrid Surgery for Obstructive Sleep Apnea - State-of-the-Art Annotation of Uvulopalatopharyngoplasty". "Scrub in" at steriletpodcast.com and on Twitter, @SterileTPodcast (twitter.com/SterileTPodcast). This podcast is a Dybas Media production. Sound effects adapted from GarageBand and sindhu.tms at https://freesound.org/people/sindhu.tms/sounds/169065/ and licensed courtesy of https://creativecommons.org/licenses/by-nc/3.0/.
On this episode of Currently Reading, Meredith and Kaytee are discussing: Bookish Moments: bookish crafting and annotation gift sets Current Reads: all the great, interesting, and/or terrible stuff we've been reading lately Deep Dive: diving into research about why reading makes us better humans The Fountain: we visit our perfect fountain to make wishes about our reading lives Show notes are time-stamped below for your convenience. Read the transcript of the episode (this link only works on the main site) . . . . 1:24 - Our Bookish Moments of the Week 2:11 - Remarkably Bright Creatures by Shelby Van Pelt 5:47 - The Gifts of Imperfection by Brene Brown 5:51 - I Guess I Haven't Learned That Yet by Shauna Niequist 5:55 - Present Over Perfect by Shauna Niequist 6:28 - Annotation gift set by Mr. Pen and Selah 8:43 - Annotation Set option 2 9:38 - Our Current Reads 9:48 - Between Flowers and Bones by Carolyn Leiloglou (Kaytee) 9:53 - CR Season 6: Episode 4 10:41 - Beneath the Swirling Sky by Carolyn Leiloglou 14:09 - Incidents Around the House by Josh Malerman (Meredith) 15:06 - Bird Box by Josh Malerman 15:10 - Daphne by Josh Malerman 15:34 - Coraline by Neil Gaiman 20:59 - Hidden Pictures by Jason Rekulak 21:01 - Baby Teeth by Zoje Stage 21:05 - We Used to Live Here by Marcus Kliewer 21:54 - Around the World in 80 Days by Jules Verne (Kaytee) 26:45 - The Four Obsessions of an Extraordinary Executive by Patrick Lencioni (Meredith) 28:05 - The Advantage by Patrick Lencioni 29:54 - The Five Temptations of a CEO by Patrick Lencioni 29:56 - The Five Dysfunctions of a Team by Patrick Lencioni 32:47 - Search by Michelle Huneven 34:13 - Unraveling by Peggy Orenstein (Kaytee) 37:38 - A Discovery of Witches by Deborah Harkness (Meredith) 40:16 - Twilight by Stephanie Meyer 41:15 - Blackwell's UK 43:09 - Outlander by Diana Gabaldon 46:02 - CR Season 1: Episode 37 47:33 - How Reading Changes Us For The Better Some Stats: 49:26 - The average reading American reads 12 books per year. 49:50 - The average American spends just $35 on books per year. 50:17 - Reading can reduce our stress levels by 68% in just six minutes. 51:04 - Reading can reduce memory decline by 30% because it activates neural pathways and can reduce the risk of Alzheimer's disease. 52:18 - Transportative fiction helps produce the most empathy in readers, but that empathy boost only lasts around 48 hours, so keep reading! 53:10 - Ghost Boys by Jewell Parker Rhodes 54:31 - Audiobooks are reading! Studies have shown that audiobooks activate the same neural pathways and cognitive benefits as print reading. 54:50 - Research shows we are less impatient with audiobooks than print. 56:02 - A Court of Thorns and Roses by Sarah J. Maas 57:13 - Meet Us At The Fountain 57:19 - I wish people would celebrate their reading in new ways. (Kaytee) 57:36 - Canon Ivy 2 Mini Photo Printer 57:44 - Storygraph 58:01 - Favorite Books of the Year print - Etsy Shop 58:31 - I wish you would give annotating books a try. (Meredith) Support Us: Become a Bookish Friend | Grab Some Merch Shop Bookshop dot org | Shop Amazon Bookish Friends Receive: The Indie Press List with a curated list of five books hand sold by the indie of the month. September's IPL comes to us from Bright Side Bookshop in Flagstaff, Arizona! Love and Chili Peppers with Kaytee and Rebekah - romance lovers get their due with this special episode focused entirely on the best selling genre fiction in the business. All Things Murderful with Meredith and Elizabeth - special content for the scary-lovers, brought to you with the behind-the-scenes insights of an independent bookseller From the Editor's Desk with Kaytee and Bunmi Ishola - a quarterly peek behind the curtain at the publishing industry The Bookish Friends Facebook Group - where you can build community with bookish friends from around the globe as well as our hosts Connect With Us: The Show: Instagram | Website | Email | Threads The Hosts and Regulars: Meredith | Kaytee | Mary | Roxanna Production and Editing: Megan Phouthavong Evans Affiliate Disclosure: All affiliate links go to Bookshop unless otherwise noted. Shopping here helps keep the lights on and benefits indie bookstores. Thanks for your support!
Send us a textIn this episode of DigiPath Digest, we review the latest AI developments in digital pathology described in the literature. I explore how AI is pushing the boundaries of metastasis detection, breast cancer treatment predictions, lung cancer research trends, and the creation of pathology foundation models. Episode Breakdown:00:00 – Welcome & Introduction00:36 – Sentinel Node Metastasis Detection: A discussion on the development of an AI model that can detect sentinel node metastasis in melanoma with accuracy comparable to that of pathologists. The model aids in distinguishing between nodal metastasis and intra-nodal nevus, which is crucial for accurate staging in melanoma patients.05:01 – Predicting Breast Cancer Treatment Response: A cross-modal AI model that integrates pathology images and ultrasound data is explored. This model is designed to predict a breast cancer patient's response to neoadjuvant chemotherapy, providing personalized insights that can guide treatment decisions.09:59 – Global Trends in AI and Lung Cancer Pathology: This section reviews a bibliometric study that analyzed global research trends in AI-based digital pathology for lung cancer over the past two decades. The study highlights the need for increased collaboration between institutions and countries to further AI advancements in this area.13:30 – Pathology Foundation Models: An in-depth look at a new foundation model in pathology, designed to generalize across various diagnostic tasks. This model shows significant promise in cancer diagnosis and prognosis prediction, outperforming traditional deep learning methods by addressing domain shifts across different datasets.20:08 – Domain Shifts in AI Models: A brief discussion on the impact of domain shifts, such as variations in staining protocols and patient populations, on the performance of AI models in pathology. Strategies for mitigating these challenges are highlighted.29:09 – Faster Annotation in Pathology: The episode concludes with a review of a study comparing manual and semi-automated annotation methods. The semi-automated approach significantly reduces the time required for annotating whole slide images, offering a more efficient solution for pathologists.Resources Mentioned:
Welcome back to Mere Mortals book reviews! Today, we're diving into Stoicism from the top with Marcus Aurelius' "Meditations." I'll share my thoughts and some interesting insights from the Penguin Classics edition. If you're new to Stoicism or curious about the Roman Emperor's personal reflections, this one's for you!(00:00) - Introduction and Overview(00:33) - Context and Historical Background(01:44) - The Purpose of "Meditations"(02:29) - Marcus Aurelius' Commitment to Stoicism(03:26) - Structure of the Penguin Classics Edition(04:51) - Notes and Annotations in the Book(05:35) - Personal Reflections and Journaling(07:03) - Timeless Wisdom and Relevance(08:31) - Marcus Aurelius' Writing Style(10:01) - Private Journal and Unique Perspective(11:17) - Reading Recommendations and Final Thoughts(12:30) - Closing Remarks and Viewer EngagementValue 4 Value Support:Boostagram: https://www.meremortalspodcast.com/supportPaypal: https://www.paypal.com/paypalme/meremortalspodcastConnect with Mere Mortals:Website: https://www.meremortalspodcast.com/Discord: https://discord.gg/jjfq9eGReUInstagram: https://www.instagram.com/meremortalspodcast/TikTok: https://www.tiktok.com/@meremortalspodcast
Because of the nature of SAM, this is more video heavy than usual. See our YouTube!Because vision is first among equals in multimodality, and yet SOTA vision language models are closed, we've always had an interest in learning what's next in vision. Our first viral episode was Segment Anything 1, and we have since covered LLaVA, IDEFICS, Adept, and Reka. But just like with Llama 3, FAIR holds a special place in our hearts as the New Kings of Open Source AI.The list of sequels better than the originals is usually very short, but SAM 2 delighted us by not only being a better image segmentation model than SAM 1, it also conclusively and inexpensively solved video segmentation in just an elegant a way as SAM 1 did for images, and releasing everything to the community as Apache 2/CC by 4.0.“In video segmentation, we observe better accuracy, using 3x fewer interactions than prior approaches. In image segmentation, our model is more accurate and 6x faster than the Segment Anything Model (SAM).”Surprisingly EfficientThe paper reports that SAM 2 was trained on 256 A100 GPUs for 108 hours (59% more than SAM 1). Taking the upper end $2 A100 cost off gpulist.ai means SAM2 cost ~$50k to train if it had an external market-rate cost - surprisingly cheap for adding video understanding!The newly released SA-V dataset is also the largest video segment dataset to date, with careful attention given to scene/object/geographical diversity, including that of annotators. In some ways, we are surprised that SOTA video segmentation can be done on only ~50,000 videos (and 640k masklet annotations). Model-in-the-loop Data Engine for Annotations and Demo-first DevelopmentSimilar to SAM 1, a 3 Phase Data Engine helped greatly in bootstrapping this dataset. As Nikhila says in the episode, the demo you see wasn't just for show, they actually used this same tool to do annotations for the model that is now demoed in the tool:“With the original SAM, we put a lot of effort in building a high-quality demo. And the other piece here is that the demo is actually the annotation tool. So we actually use the demo as a way to improve our annotation tool. And so then it becomes very natural to invest in building a good demo because it speeds up your annotation. and improve the data quality, and that will improve the model quality. With this approach, we found it to be really successful.”An incredible 90% speedup in annotation happened due to this virtuous cycle which helped SA-V reach this incredible scale.Building the demo also helped the team live the context that their own downstream users, like Roboflow, would experience, and forced them to make choices accordingly.As Nikhila says:“It's a really encouraging trend for not thinking about only the new model capability, but what sort of applications folks want to build with models as a result of that downstream.I think it also really forces you to think about many things that you might postpone. For example, efficiency. For a good demo experience, making it real time is super important. No one wants to wait. And so it really forces you to think about these things much sooner and actually makes us think about what kind of image encoder we want to use or other things. hardware efficiency improvements. So those kind of things, I think, become a first-class citizen when you put the demo first.”Indeed, the team swapped out standard ViT-H Vision Transformers for Hiera (Hierarchical) Vision Transformers as a result of efficiency considerations.Memory AttentionSpeaking of architecture, the model design is probably the sleeper hit of a project filled with hits. The team adapted SAM 1 to video by adding streaming memory for real-time video processing:Specifically adding memory attention, memory encoder, and memory bank, which surprisingly ablated better than more intuitive but complex architectures like Gated Recurrent Units.One has to wonder if streaming memory can be added to pure language models with a similar approach… (pls comment if there's an obvious one we haven't come across yet!)Video PodcastTune in to Latent Space TV for the video demos mentioned in this video podcast!Timestamps* [00:00:00] The Rise of SAM by Udio (David Ding Edit)* [00:03:07] Introducing Nikhila* [00:06:38] The Impact of SAM 1 in 2023* [00:12:15] Do People Finetune SAM?* [00:16:05] Video Demo of SAM* [00:20:01] Why the Demo is so Important* [00:23:23] SAM 1 vs SAM 2 Architecture* [00:26:46] Video Demo of SAM on Roboflow* [00:32:44] Extending SAM 2 with other models* [00:35:00] Limitations of SAM: Screenshots* [00:38:56] SAM 2 Paper* [00:39:15] SA-V Dataset and SAM Data Engine* [00:43:15] Memory Attention to solve Video* [00:47:24] "Context Length" in Memory Attention* [00:48:17] Object Tracking* [00:50:52] The Future of FAIR* [00:52:23] CVPR, Trends in Vision* [01:02:04] Calls to ActionTranscript[00:00:00] [music intro][00:02:11] AI Charlie: Happy Yoga! This is your AI co host Charlie. Thank you for all the love for our special 1 million downloads Wins of AI Winter episode last week, especially Sam, Archie, Trellis, Morgan, Shrey, Han, and more. For this episode, we have to go all the way back to the first viral episode of the podcast Segment Anything Model and the Hard Problems of Computer Vision, which we discussed with Joseph Nelson of Roboflow.[00:02:39] AI Charlie: Since Meta released SAM 2 last week, we are delighted to welcome Joseph back as our fourth guest co host to chat with Nikhila Ravi, Research Engineering Manager at Facebook AI Research and lead author of SAM 2. Just like our SAM 1 podcast, this is a multimodal pod because of the vision element, so we definitely encourage you to hop over to our YouTube at least for the demos, if not our faces.[00:03:04] AI Charlie: Watch out and take care.[00:03:10] Introducing Nikhila[00:03:10] swyx: Welcome to the latest podcast. I'm delighted to do segment anything to our first, one of our very first viral podcasts was segment anything one with Joseph. Welcome back. Thanks so much. And this time we are joined by the lead author of Segment Anything 2, Nikki Ravi, welcome.[00:03:25] Nikhila Ravi: Thank you. Thanks for having me.[00:03:26] swyx: There's a whole story that we can refer people back to episode of the podcast way back when for the story of Segment Anything, but I think we're interested in just introducing you as a researcher, as a, on the human side what was your path into AI research? Why, you know, why did you choose computer vision coming out of your specialization at Cambridge?[00:03:46] Nikhila Ravi: So I did my undergraduate. Degree in engineering at Cambridge university. The engineering program is very general. So first couple of years, you sort of study everything from mechanical engineering to fluid mechanics, structural mechanics, material science, and also computer science.[00:04:04] Nikhila Ravi: Towards the end of my degree, I started taking more classes in machine learning and computational neuroscience, and I really enjoyed it. And actually after graduating from undergrad, I had a place at Oxford to study medicine. And so I was. Initially planning on becoming a doctor, had everything planned and then decided to take a gap year after finishing undergrad.[00:04:28] Nikhila Ravi: And actually that was around the time that sort of deep learning was emerging. And in my machine learning class in undergrad, I remember one day our professor came in and that was when Google acquired DeepMind. And so that became like a huge thing. We talked about it for the whole class. It kind of really stuck.[00:04:48] Nikhila Ravi: And I was kicked off thinking about, okay, maybe I want to try something different other than medicine. Maybe this is a different path I want to take. And then in the gap year, I did a bunch of coding, worked on a number of projects. Did some sort of freelance contracting work. And then I got a scholarship to come and study in America.[00:05:06] Nikhila Ravi: So I went to Harvard for a year, took a bunch of computer science classes at Harvard and MIT, worked on a number of AI projects, especially in computer vision. I really, really enjoyed working in computer vision. I applied to Facebook and got this job at Facebook, and I've now at Facebook at the time, now Meta, and I've been here for seven years, so very circuitous path, probably not a very unconventional, I didn't do a PhD, I'm not like a research, typical research scientist, definitely came from more of an engineering background, but since being at Meta, Have had amazing opportunities to work across so many different interesting problems in computer vision from 3D computer vision.[00:05:50] Nikhila Ravi: How can you go from images of objects to 3D structures and then going back to 2D computer vision and actually understanding the objects and the pixels and the images themselves. So it's been a very interesting journey over the past seven years.[00:06:05] swyx: It's weird because like, I guess with segment anything too, it's like 4D because you solve time, you know, you started with 3D and now you're solving the 4D.[00:06:14] Nikhila Ravi: Yeah, it's just going from 3D to images to video. It's really covering the full spectrum. And actually, one of the nice things has been, so I think I mentioned I, Wanted to become a doctor, but actually Sam is having so much impact in medicine, probably more than I could have ever had as a doctor myself. So I think, you know, hopefully Sam too can also have a similar sort of impact in medicine and other fields.[00:06:39] The Impact of SAM 1 in 2023[00:06:39] swyx: Yeah. I want to give Joseph a chance to comment. Does that also mirror your, we know your story about going into, into vision, but like in the past year, since we did our podcast on Sam what's been the impact that you've seen?[00:06:51] Joseph Nelson: Segment anything. Set a new standard in computer vision, you know recapping from from the first release to present Sam introduces the ability for models to near zero shot meaning without any training identify kind of perfect polygons and outlines of items and objects inside images and that capability previously required a Lots of manual labeling, lots of manual preparation, clicking very meticulously to create outlines of individuals and people.[00:07:25] Joseph Nelson: And there were some models that attempted to do zero shot segmentation. of items inside images, though none were as high quality as segment anything. And with the introduction of segment anything, you can pass an image with SAM1, SAM2 videos as well, and get perfect pixel perfect outlines of most everything inside the images.[00:07:52] Joseph Nelson: Now there are some edge cases across domains and Similar to the human eye, sometimes you need to say, like, which item maybe you most care about for the downstream task and problem you're working on. Though, SAM has accelerated the rate at which developers are able to use computer vision in production applications.[00:08:13] Joseph Nelson: So, at RoboFlow, we were very quick to enable the community of computer vision developers and engineers to use SAM and apply it to their problems. The principle ways of using SAM, you could kind of use SAM as is to like pass an image and receive back masks. Another use case for SAM is in preparation of data for other types of problems.[00:08:37] Joseph Nelson: So, for example, in the medical domain, let's say that you're working on a problem where you have a bunch of images from a wet lab experiment. And from each of those images, you need to count the presence of a particular protein that reacts to some experiment. To count all the individual protein reactions, You can go in and lab assistants to this day will still like kind of individually count and say what are the presence of all those proteins.[00:09:07] Joseph Nelson: With Segment Anything, it's able to identify all of those individual items correctly. But often you may need to also add like a class name to what the protein is. Or you may need to say, hey, like, I care about the protein portion of this. I don't care about the rest of the portion of this in the image.[00:09:26] Joseph Nelson: And, or what it encourages and asks for the user to do is to provide some visual prompting to say, hey, which part, like, Sam says, hey, I can find segments of anything, but which segments do you care about? And so you can do visual prompting, which is kind of a new primitive that Sam introduced. And so at RoboFlow, we have one portion of our tool stack enables users to very quickly label data.[00:09:48] Joseph Nelson: With segment anything, Sam can already provide, hey, here's where I see the outlines of objects. Or a user can click to prompt to say, Hey, here's where the outlines of objects matter. And I recently pulled statistics from the usage of SAM in RoboFlow over the course of the last year. And users have labeled about 49 million images using segment anything on the hosted side of the RoboFlow platform.[00:10:12] Joseph Nelson: And that's like 5 million in the last 30 days alone. And of those images, We did kind of like a rough bafka napkin calculation of like how much time that has saved. Because, again, the alternative is you're clicking individual points to create a polygon, and with SAM you just click once and it guesses where the polygon is.[00:10:32] Joseph Nelson: And I'm sure in a bit we can maybe screen share and show some examples of what this experience is like. And in that time estimation, it's like, On average saves, you know, maybe a dozen or so seconds. And we estimate that this is probably saved on the order of magnitude of 35 years of time for users.[00:10:53] Nikhila Ravi: That's incredible.[00:10:54] Joseph Nelson: So, I mean, basically like in the first, the first year of a model being available, not only can you say, Hey, I'm just going to go use this model, those numbers that like 49 million images. is an estimate directly related to just the hosted side. So imagine all of the users that are self hosting or using SAM for robotics applications or out in the field or offline where it's not even, like, the time or the image counts are tabulated.[00:11:20] Joseph Nelson: And we're probably talking about, you know, just a fraction of the amount of value that's actually being produced for a number of downstream tasks. So to say that the impact has been You know, people use terms like game changing and these sorts of things. It has changed the industry. It's set a new standard.[00:11:36] Joseph Nelson: And with the release of SAM 2, I think we're about to see an acceleration of those capabilities for a lot of reasons.[00:11:42] Nikhila Ravi: That's really great to hear. I think one of the, really SAM 1 was. How many fields actually rely on manual segmentation? I think we're not really exposed to that. Maybe you are at Roboflow because you get to see all the users of these tools.[00:11:57] Nikhila Ravi: But for me, it was, you know, people working on understanding coral reef bleaching or farmers counting their cows and so many different applications that as a researcher. You never get exposed to, but you can have impact towards. So I think that was really awesome to hear.[00:12:15] Do People Finetune SAM?[00:12:15] swyx: So as sort of audience surrogate, who knows less than the two of you, I'm going to ask a really dumb question maybe, but is everyone using stock, a segment, anything?[00:12:23] swyx: Are they fine tuning for the medical domain? Like how on earth could it work for the medical field without fine tuning, right? Like, is that a thing?[00:12:32] Nikhila Ravi: So I mean, I can give a quick perspective from the research side. So one of the things, design decisions we made in SAM was to not have class labels. And so all the data is annotated in a class agnostic way.[00:12:48] Nikhila Ravi: So anything that has a boundary, we consider to be an object. So for example, in any image, there's lots of small objects. We might not know what the name of them are, but they're If you can draw a boundary around it, so you can imagine that we have 11 million images in the SA 1B dataset, we annotated all the objects, there's many, many small objects.[00:13:12] Nikhila Ravi: And so if you think about cells, they're also kind of small objects, there's probably things in the training data. That looked like it, but we didn't have to label it. And so that means that even when you use SAM for applications that it wasn't really trained for, because we didn't restrict it to a certain set of categories, you can actually use it out of the box without custom adaptation.[00:13:35] Nikhila Ravi: But having said that, there's probably certain domains where you need some expertise in order to be able to segment something properly. And for those use cases, Having some extra fine tuning data would probably help, and we've sort of seen that there's some papers that have come out that do this, and, you know, we'd love to hear, Joseph, how people are collecting data with SAM and fine tuning for their use cases.[00:13:59] Joseph Nelson: Once SAM came out, there were adaptations that said, could we use SAM to be, you know, like, efficient SAM? Like, basically take SAM and maybe accelerate it. And then there were domain adapted SAMs, like CellSAM, for example, out of the UC system. Now, what's interesting is, there's, like, adapting SAM to a domain, there's kind of two ways by which that's done.[00:14:21] Joseph Nelson: One is, as you mentioned, like, potentially SAM doesn't have a good concept of The objects of interest. And so you need to do domain adaptation and increase the accuracy for zero shot prediction. The second way though, is it's not fine tuning. It's actually just prompting. It's just guiding the model existing knowledge.[00:14:42] Joseph Nelson: to say which segments you care about. And both those are actually kind of equally important on the application side. You need to, like, a priori ensure that the objects of interest can be correctly segmented and maybe collect data to do that. But even if you had, like, a perfect SAM, like an omniscient SAM that could see every segment in every domain with all pixels perfectly outlined, in production, you would still need some way to Almost like signal to the model what you care about like to paint this picture if you are like a retailer and you are providing Photos of models wearing your clothing on your retail site You may care about you know only the shirt and Sam by default might segment the full person And so there's you know visual prompting that you can do to ensure that you only outline Maybe the shirt for the purposes of swapping in and out different shirts for displaying a given model on a retail page You And so I think what's interesting is that's where, like I wouldn't call it domain adaptation, but that's where, like, when you apply to industry, like, one thing that's particularly important with tooling and enabling SAM to reach its full potential.[00:15:51] swyx: That's really encouraging to hear. I should also think, like, you know, the last time we talked about this, we wanted to, the very natural addition on the class labeling side is the grounding Dino work, right? So I think people, built a grounding SAM and all the other extensions.[00:16:05] Video Demo of SAM[00:16:05] swyx: I think it's, it's probably a good time to cut to a quick demo of SAM2 for people who are, who are tuning in for SAM2 and who better to demo SAM2 than Nikki.[00:16:15] Nikhila Ravi: Sure. So I'll try to narrate what I'm what I'm doing. So audio listeners can also understand. So we have a web demo where anyone can try SAM2 on a video. Here we have a video of someone kicking a football, and I'm going to click on the football to select the object in the first frame. But you can actually select the object in any frame of the video, and this will work.[00:16:40] Nikhila Ravi: The next step is to hit track. So the model's now tracking this in real time. We don't save any of this, it's all running in real time. And now you can see the ball has been tracked throughout the entire video. There's even like a little bit of a challenging case here where the shoe covers the football.[00:16:59] Nikhila Ravi: And actually, you know, the model makes a little bit of a mistake, but that's okay. Because we can actually, here, the model makes a little bit of a mistake here. But you know, we can actually add a refinement click. You can add negative clicks until we get the mask that we want on this frame. And then you can hit track again, and the model will track the object, taking into account the additional information I've provided at that frame.[00:17:25] Nikhila Ravi: We've also added a couple of other fun things you can do on top of the track, like add effects. We can add you know, foreground effects, background effects. And these are just ways of showing how we can use the output from SAM2 as part of other tools like video editing tools. Other systems, so this is just a preview of what you can do with SAM2, but the really cool use cases are places where we might not have even imagined SAM2 being useful.[00:17:54] Nikhila Ravi: So we have a number of examples of things you might want to use it for. There's like underwater videos that it works actually really well for even though we, models never really seen an octopus before and octopus have a lot of moving parts that SAM2 can actually quite effectively. Keep track of all the different tentacles and we can probably see it more clearly if I desaturate the background.[00:18:18] Nikhila Ravi: We can see that actually the tracking of all the different tentacles is Quite accurate. Another challenge with video is that objects can actually become occluded. They can disappear from view and reappear. And a really fun example here is the shuffling cup game, which many of you might have seen. And so here I can click on the ball in the first frame.[00:18:41] Nikhila Ravi: I can also, You know, click on a different cup. And so here, the additional challenge is that there's three cups that look exactly the same. And then there's the ball that will get occluded by the cup. So the ball's no longer visible, the cups are all moving around, they all look the same. But the model actually keeps track of the cup that we selected.[00:19:02] Nikhila Ravi: And, as you can see at the end, here I'll jump to the end so you can see. It actually finds the cup again. I wanted to point out a couple of fun demo UX features that we added that actually really helped with this. So if you can see at the bottom, there's these swim lanes and then the swim lanes, actually the thickness of the swim lane tells you if the object's visible or not.[00:19:22] Nikhila Ravi: So at the beginning, the object's visible,[00:19:25] swyx: the object[00:19:26] Nikhila Ravi: disappears, and then the object comes back. So you can actually visually tell. When the object's being occluded and when it's not, and so it's a nice way of like, knowing if you need to go in and fix the model prediction or not. And so these are some of the UX innovations that we came up with, as well as the model innovations.[00:19:46] Joseph Nelson: One thing that I think is really notable here, there's two things. One is that like, I'd love to have a little bit of a discussion about how the models keeping track of the embedded scene to keep track of the ball and the cup in different places. Put a pause on that for a second.[00:19:59] Why the Demo is so Important[00:19:59] Joseph Nelson: One thing that Meta has put an emphasis on here in a much greater degree than other model releases is the demo experience of recognizing that in addition to having a model that can do zero shot segmentation, you've created a web experience that allows folks to kind of experience both the video effects but the types of UX innovations that encourage usage and adoption.[00:20:23] Joseph Nelson: It's actually kind of reminiscent of The underlying technology of ChatGPT was available prior to the web experience of ChatGPT. Can you talk a bit about why that was a consideration to your team and how you thought about the creation of The demo experience in tandem with training and releasing a new model.[00:20:41] Nikhila Ravi: Yeah, absolutely. I think that's a really great example of how, you know, Chad, GPT was really more of a UX innovation. Obviously it was like a number of research innovations that helped to get to this point. But as you said, like the underlying technology was around for a while. And, you know, putting this UX around as a chat interface helped tremendously with the.[00:21:03] Nikhila Ravi: Adoption and people understanding how it could be useful for real world use cases. And in computer vision, especially, it's so visual. The best way to show how these models work. Is by trying it on your own image or your own video with the original SAM, we put a lot of effort in building like a high quality demo.[00:21:23] Nikhila Ravi: And the other piece here is that the demo is actually the annotation tool. So we actually. Use the demo as a way to improve our annotation tool. And so then it becomes very natural to invest in building a good demo because it speeds up your annotation and improves the data quality and that will improve the model quality.[00:21:43] Nikhila Ravi: With this approach, we found it to be really successful. And obviously externally, people really liked being able to try it. I think, you know, people in fields outside of machine learning would never have tried SAM if we didn't have that demo. And I think that definitely led to a lot of the adoption in, like, diverse fields.[00:22:05] Nikhila Ravi: And so because we saw that with SAM 2, like, the demo was a priority first class citizen from day one. And so we really invested in making that. And I think with SAM2 as well, we wanted to have like a step change in the demo experience. Interactive video segmentation, I think that experience is something that maybe has not had much thought given to it.[00:22:27] Nikhila Ravi: And we really wanted to be like, okay, if we are to design a step changing video segmentation experience, what would that look like? And that really did influence our model. And annotation design as well.[00:22:40] Joseph Nelson: It's a really encouraging trend for not thinking about only the new model capability, but what sort of applications folks want to build with models as a result of that downstream.[00:22:49] Nikhila Ravi: I think it also really forces you to think about many things that you might postpone, for example, efficiency.[00:22:55] Joseph Nelson: Yes.[00:22:55] Nikhila Ravi: For a good demo experience. Making it real time is super important. No one wants to wait. And so it really forces you to think about these things much sooner and actually makes us think about how to, what kind of image encoder we want to use or like other hardware efficiency improvements.[00:23:13] Nikhila Ravi: So those kinds of things, I think, become a first class citizen when you put the demo first.[00:23:19] SAM 1 vs SAM 2 Architecture[00:23:19] Joseph Nelson: That's one thing I was going to ask about, and this is related to the architecture change. So SAM1 and the SAM1 demo experience. You have the encoder that's creating the embeddings of all the potential spaces.[00:23:31] Joseph Nelson: That needs to be run on a GPU. That's a relatively intensive operation. But then the query of those embeddings can be run independently and on a cheaper process. So in the SAM1 demo, the way that it was structured, and also this is the way that we have our SAM tool structured in Robloflow as well, is images go to a GPU to get all the SAM based embeddings.[00:23:53] Joseph Nelson: But then for querying those embeddings, we do that client side, in the browser, so that the user can very quickly, you know, you can move your mouse over and you get the proposed candidate masks that Sam found for that region of the image. In SAM 2 you dropped that in the web demo. And I think that's because you made some notable improvements to the rate at which encoding happens.[00:24:16] Joseph Nelson: Can you talk a bit about what led to those speed increases and, again, how that interplays with providing a fast encryption? user experience for interacting with the model.[00:24:29] Nikhila Ravi: Yeah. So the SAM2 web demo is primarily focused on video. We, we decided to just keep it simple and focus on video and on GitHub, we have a Colab notebook that shows how to run SAM2 on images.[00:24:41] Nikhila Ravi: So if you're interested in using, replacing SAM with SAM2 for images, check out GitHub, but on the SAM2 demo, it's not as straightforward to adopt the same architecture as SAM. For video, because we can't send the per frame image embeddings for an entire video back to the front end. In SAM, each frame embedding was like four megabytes, but if you have a long video and that's like per frame, it would become impossible to send that back to the front end.[00:25:11] Nikhila Ravi: So, SAM 2 actually, in terms of the architecture details, I was actually just looking at this earlier, but SAM1 model was around 630 million parameters. It's a fraction of the size of these large language models, but very small. Actually, SAM2, the largest model, is around 224 million parameters. So it's actually One third the size of the SAM original model.[00:25:38] Nikhila Ravi: So we changed the imaging coder from A-V-I-T-H and SAM to a higher model, which has also developed by by meta. So that definitely was something that helped. And in terms of the efficiency compared to sam, so if we were to run SAM per frame on a video or run SAM two, it's around six times faster to run SAM two versus run SAM per frame.[00:26:03] Nikhila Ravi: A number of things improved the efficiency of SAM2 such that we were actually able to run this entirely on the server and not have any component in the front end. But I am very curious to see who puts this on device, like I'm pretty sure soon we'll see like an on device SAM2 or, you know, maybe even running in the browser or something, so.[00:26:25] Nikhila Ravi: I think that could definitely unlock some of these edge use cases that we were able to make a compelling web demo without having to do that.[00:26:34] swyx: Hugging face is probably already working on Transformers. js version of it, but totally makes sense. I want to talk about more about things from the paper, but I think we're still in this sort of demo section.[00:26:42] Video Demo of SAM on Roboflow[00:26:42] swyx: And so I want to hand it to Joseph for his demo to see what the RoboFlow site looks like.[00:26:47] Joseph Nelson: So I can, I can give some context into one key area that Nicola, you mentioned earlier, which is. Sam has made the decision, both Sam 1 and Sam 2, to be class agnostic in terms of its predictions. And that, you then have the ability to have a generalizable, model for zero shot capability.[00:27:05] Joseph Nelson: However, in a lot of domain applications, you do want the class wise name. And so a lot of the challenge can be adding that class wise name for the, at least the annotation to an experience that we've created. That's one of the key considerations. So I will similarly Share my screen and show an example.[00:27:27] Joseph Nelson: Here, I have a bunch of images, and there's a number of ways that I could annotate things, like I could prompt a large multimodal model with like grounding capabilities, you know, you could outsource it, or I can do manual labeling. And with the manual labeling, this is where we make use of models like segment anything.[00:27:45] Joseph Nelson: to propose candidate masks and make it faster. So we have, you know, this annotation pane and what we call the smart poly tool, which is powered by Segment Anything. This is currently Segment Anything 1. We're accelerating and seeing improvements from similar to what the paper shows of Segment Anything 2 performed better on E3.[00:28:06] Joseph Nelson: Images as well as video, but with a segment, anything I'm able to basically prompt regions of my image of interest. So for example, if like, I wanted to say, I want to like add the drum set. You'll see here that like, the original candidate proposal is just the base drum, but let's say I wanted the whole drum set.[00:28:26] Joseph Nelson: So the UX primitive of being able to add and subtract candidate regions of interest is really intuitive here. And now, great, I have this outline, but in fact what I want is, I want to name that as a class. Because maybe for the model that I'm building, I want to build like a task specific model, you know, like an object detection model or an instant segmentation model.[00:28:50] Joseph Nelson: Or, you know, maybe I'm even using like a multimodal model and I want that multimodal model to refer to regions of interest in the images as a specific thing. And so I think what's, you know, really powerful is, of course, like, I get this really rich zero shot prediction. And here we have our friend Rick.[00:29:10] Joseph Nelson: So I get this really rich candidate set of predictions. But then by adding the class wise label, I can, you know, very quickly make sure that any downstream tasks are aware not just of the segment, but also of the, what is inside that segment. Which actually takes me to A separate point of something that I predict that's probably going to happen and Nikhil, I'm actually kind of interested why maybe your team made a conscious decision to not do this initially with SAM2.[00:29:40] Joseph Nelson: There's been an emergent set of models that are also adding open text prompting capabilities to grounding models. So for example, like you've seen models like Grounding Dino or Owlvit, which, you know, you can do. Even image to image or text to image based prompting to find regions of interest. And maybe maybe I can actually give an example of that even in the context of this same data.[00:30:05] Joseph Nelson: So if I wanted to try out, you know, grounding dino on this same set of images, I could try out, you know, prompting grounding dino for a set of different classes. And what's notable is let's do, I don't know, let's prompt for person and we'll prompt for person and prompt for I don't know, microphone.[00:30:26] Joseph Nelson: NLASC or microphone. Here I can text prompt the image and then the understanding, in this case Grounding Dino's understanding, of where people are in this image allows me to create, in this case, bounding boxes, but, you know, soon you can do segmentations or in tandem with SAM do segmentations. And, you know, we've already seen applications of using SAM2 in tandem with models like Grounding Dino or Florence 2.[00:30:54] Joseph Nelson: So that people can basically text prompt and then get the benefits of the zero shot segmentation at the same time as getting the open form querying. And in doing so, you know, we maintain a framework called like autodistill so like folks can very quickly, you know, bring some images and then using autodistill to find some ontology and then prompt and say what you want from that ontology.[00:31:19] Nikhila Ravi: So you already do this for video as well?[00:31:21] Joseph Nelson: You can apply videos or groups of images, yes. So this is using a project called Autodistill. And the concept of Autodistill is, use a base model, like a big base model, which could be like SAM or Grounding Dino, and then you pass a directory of images, which also could be video, broken into individual frames, and you pass an ontology as well.[00:31:43] Joseph Nelson: So an example I was just showing was like the hello world we have, which is like a shipping container. And then the combination of the grounding capabilities of, in the example I was showing, Florence 2 plus SAM, looks for the concept of container, and then SAM does the rich segmentation of turning that concept of container into the candidate proposal of the region, so that a user could just say, hey, I want all the shipping containers, run this across a bunch of images or video frames, And then get back the class wise labels plus the regions of interest.[00:32:17] Joseph Nelson: And this feels like a natural extension. And in fact, like the open form grounding capabilities between SAM1 and SAM2 became something the field was broadly doing. So I'm curious, like, from your perspective, one of the things I thought maybe SAM2 would do is actually add this capability natively. So I'm curious to hear, like, the conscious decision to say, hey, we want to continue to be class agnostic.[00:32:39] Extending SAM 2 with other models[00:32:39] Joseph Nelson: We don't want to add yet maybe open form text prompting as a part of finding the segments and parts of images. And I'd love to hear about like the decision to think about it that way. And if you are encouraged or if you want kind of like what's happening here where people are naturally combining these capabilities as something that you would expect and encourage to happen despite not having it.[00:33:00] Joseph Nelson: In the base model itself.[00:33:02] Nikhila Ravi: Yeah, it's a great question. So I think it's really cool that the community is taking SAM and taking SAM 2 and building on top of it and coming up with cool applications. We love to see that. That's exactly why we open source our work. And then in terms of why we didn't put it into SAM 2, so as you've probably seen with SAM and SAM 2, it's a fairly narrow problem.[00:33:25] Nikhila Ravi: But we really tried to make it a step change in the capability. And so with each version, we are trying to limit the focus on one thing that we can know we can do really well. And in this case, like the first SAM, it was class agnostic segmentation, but can we do it so well that it's effectively solved?[00:33:47] Nikhila Ravi: And similarly, can we do that same thing, but with Video segmentation. So one step at a time, we are working on each of these problems one at a time so that we can actually deliver something that's really world class and step changing.[00:34:03] Joseph Nelson: So does that mean SAM 3 will have the text prompting? Problem is like the next challenge.[00:34:09] Nikhila Ravi: Who knows, who knows? Maybe the community will, will we'll build that too. So[00:34:15] Joseph Nelson: it makes sense to like very narrowly do something very well. And that's, I think, proven to be well accomplished.[00:34:21] Nikhila Ravi: It's like taking the, the, both the data, the model and the demo, and how can we push all three towards solving one thing really well?[00:34:30] Nikhila Ravi: So we found that. That's like a good recipe and that's what we've limited the focus of these, of each of these models.[00:34:38] swyx: This development reminds me of how, you know, when you do, and you break out the interpretability of ConvNets and you can see like, Oh, this is the edge detection one. I feel like SAM is the edge detection version equivalent.[00:34:51] swyx: And then you build up to whatever the next feature is on top of that.[00:34:54] Limitations of SAM: Screenshots[00:34:54] Joseph Nelson: Can I bring up one? Limitation of SAM. So like we've like even SAM one, SAM two, and the monitor is released at 4 PM Pacific on Monday. We're recording this on 11 AM Pacific on, on, on Thursday. So the, it's very fresh for a lot of the capabilities and.[00:35:09] Joseph Nelson: It is so clear that it is a stepwise change in the capability that, Nikhila, you mentioned your team wants to do, which is extend SAM's zero shot class agnostic capability to video, like, A plus, kind of mission accomplished. One thing that's interesting is finding, like, domain problems where there might be still domain applicability and domain adaptation that is available.[00:35:32] Joseph Nelson: One benchmark that we introduced at CBPR is this thing called RF100, which is like, seven different domain type problems that the industry commonly is working on in vision, like underwater document processing, aerial examples, medicine examples. And one place where interestingly segment anything maybe less performant than other models is handling screenshots.[00:35:57] Joseph Nelson: For example, like a lot of folks that are building agents to interact with the web are particularly interested in that challenge of given a screenshot of a computer, what are all the buttons. And how could I autonomously navigate and prompt and tell it to click? And I can show an example of like maybe what, how like Sam kind of performs on this challenge just to outline some of the context of this problem.[00:36:23] Joseph Nelson: But I'm curious like how you think about limitations like this and what you would expect to want to be the case. So here I just have a notebook where I run Sam on the source image on the left. Or the source image on the left and then Sam output is on the right. And this is just a screenshot of, of a website where we just grab like the top 100 websites by traffic and grab screenshots from them.[00:36:42] Joseph Nelson: One example of a place where I could see the community improving on Sam, and I'm curious how you think about this challenge and maybe why Sam is less well adapted for this type of problem. Is processing screenshots. So I'll share my screen to give an example for, for viewers that are participating here, you see like an example, a screenshot of a website on the left, and then right is SAM two running on that image.[00:37:06] Joseph Nelson: And in the context of agents, folks usually want to have like, Hey, tell me all of the buttons that a, an agent could press. Tell me like maybe the headlines of the articles tell me the individual images and Sam two behaves perhaps predictably, where it outlines like people in the images and like some of like the, the screen text.[00:37:22] Joseph Nelson: I'm curious, like, how you think about a challenge like this for a model that sees everything in the world, what about handling digital contexts? And Why maybe it could perform better here and how you would expect to see improvement for domains that might have been out of distribution from the training data?[00:37:40] Nikhila Ravi: Yeah, this is a good question. So fair, we don't really build with a specific use case in mind. We try to build like these foundational models that can be applied to lots of different use cases out of the box. So I think in this kind of example, potentially people might want to annotate some data.[00:37:59] Nikhila Ravi: Fine tune on top of what we release. I think we probably won't build things that are very custom for different use cases. I think that's not a direction we'll go in, but as you said, like the model is an annotation tool to improve the model. And so I think that's definitely the approach we want to take is we provide the tools for you to improve the model as well as the model itself.[00:38:27] Joseph Nelson: That makes sense. Focus on like as many. Multi or zero shot problems and then allow the community to pick up the torch for domain adaptation.[00:38:34] Nikhila Ravi: Yeah, absolutely. Like, we can't solve all the problems ourselves. Like, we can't solve all the different domains. But if we can provide a sort of base hammer tool, and then people can apply it to all their different problems.[00:38:48] SAM 2 Paper[00:38:48] swyx: If you don't mind, I guess we want to transition to a little bit on like asking more questions about the paper.[00:38:53] Udio AI: Sure.[00:38:54] swyx: There's a lot in here. I love the transparency from Meta recently with like LLAMA 3 last week and then, and was it last week? Maybe, maybe a little bit less than last week. But just like just really, really well written and a lot of disclosures, including the data set as well.[00:39:08] SA-V Dataset and SAM Data Engine[00:39:08] swyx: I think the top question that people had on the data set, you know, you release a diverse videos and there was, there's a lot of discussion about the data engine as well, which I really love. And I think it's innovative if you wanted. I think the top question is like, how do you decide the size of data set?[00:39:22] swyx: You know, what were you constrained by? People are asking about scaling laws. You had some ablations, but as a research manager for this whole thing, like how do you decide what you need?[00:39:32] Nikhila Ravi: Yeah. I mean, it's a great question. I think it's, as with all papers, you write them at the end of the project, so we can put these nice plots at the end, but going into it, I think, you know, the data engine design really follows.[00:39:47] Nikhila Ravi: So, this is sort of the model design, how we thought about the task, how we thought of the model capabilities. You can really see it's reflected in the different phases of the data engine. We started with just SAM, we apply SAM per frame. That's like the most basic way of extending SAM to video. Then the most obvious thing to do is to take the output masks from SAM and then provide it as input into a video object segmentation model that takes the mask as the first frame input.[00:40:19] Nikhila Ravi: And that's exactly what we did. We had SAM plus a version of SAM2 that only had mask as input. And then in the last phase, we got rid of SAM entirely and just had this one unified model that can do both image. And video segmentation. And I can do everything in just one model. And we found that, you know, going from each phase, it both improved the efficiency and it improved the data quality.[00:40:46] Nikhila Ravi: And in particular, when you get rid of this two part model, one of the advantages is that when you make refinement clicks, so, You prompt the model in one frame to select an object, then you propagate those predictions to all the other frames of the video to track the object. But if the model makes a mistake and you want to correct it, when you have this unified model, you only need to provide refinement clicks.[00:41:14] Nikhila Ravi: So you can provide maybe a negative click to remove a region or a positive click to add a region. But if you had this decoupled model, you would have to Delete that frame prediction and re annotate from scratch. And so you can imagine for more complex objects, this is actually adding like a lot of extra time to redefine that object every time you want to make a correction.[00:41:39] Nikhila Ravi: So both the data and the data engine phases really follow, like how we thought about the model design and the evolution of the capabilities, because it really helped us to do that. improve the data quality and the annotation efficiency as well.[00:41:54] swyx: Yeah, you had a really nice table with like time taken to annotate and it was just going down and down.[00:41:58] swyx: I think it was like down by like 90 percent by the time you hit stage[00:42:02] Joseph Nelson: three, which is kind of cool. We joke that when SAM 1 came out at RoboFlow, we're like, was this purpose built for our software? Like you have like the embedding, you have the embedding take like a big model and the querying of the embeddings A smaller model that happens in browser, which felt remarkably aligned.[00:42:18] Joseph Nelson: Now hearing you talk about how you think about building models with a demo in mind, it makes sense. Like, you're thinking about the ways that folks downstream are going to be consuming and creating value. So, what felt like maybe a coincidence was perhaps a deliberate choice by Meta to take into account how industry is going to take Seminal advances and apply them.[00:42:36] Nikhila Ravi: Yeah. And it's not just humans. Like it could also be a model that outputs boxes that then get fed into this model. So really thinking about this as a component that could be used by a human or as a component, as part of a, of a larger AI system. And that has, you know, a number of design requirements. It needs to be promptable.[00:42:56] Nikhila Ravi: It needs to be, have the zero shot generalization capability. We, you know, need it to be real time and. Those requirements really are very core to how we think about these models.[00:43:08] Memory Attention to solve Video[00:43:08] swyx: I cannot end this podcast without talking about the architecture, because this is your, effectively the sort of research level, architecture level innovation that enabled what I've been calling object permanence for SAM.[00:43:22] swyx: And it's memory retention. What was the inspiration going into it? And you know, what did you find?[00:43:27] Nikhila Ravi: Yeah, so at a high level, the way we think about extending SAM to video is that an image is just a special case of a video that just has one frame. With that idea in mind, we can extend the SAM architecture to be able to support segmentation across videos.[00:43:45] Nikhila Ravi: So this is a quick video that shows how this works. So SAM architecture, we have the image encoder, we have a prompt encoder, we have a mask decoder. You can click on an image. And that basically is a prompt, we use that prompt along with the image embedding to make a mask prediction for that image. Going to SAM2, we can also apply SAM2 to images because we can, you know, as I said, treat an image as a video with a single frame.[00:44:15] Nikhila Ravi: And so when we, in the SAM2 architecture, we introduce this new memory mechanism that consists of three main components. There's memory attention, there's a memory encoder, and then there's a memory bank. And when we apply SAM2 to images, these are effectively not used. And the architecture just collapses down to the original SAM architecture.[00:44:35] Nikhila Ravi: But when we do apply this to video, the memory components become really useful because they provide the context of the target object from Other frames. And so this could be from past frames. It can be from, there's two types of memory. So there's like the condition, conditional frames or the prompted frames, which are basically the frames at which a user or a model provides input like clicks.[00:45:01] Nikhila Ravi: And then there's like the surrounding frames. And say we use six frames around the current frame as memory of the object. So there's, there's those, those, both those types of memory that we use to make the prediction. Going into a little bit more detail about that, there's like two kinds of memory that we use.[00:45:18] Nikhila Ravi: So one is like spatial memory. So it's like this high resolution memory that captures the spatial details. And then we also have this like longer term object pointer memory that captures some of the sort of higher level concepts. And I think Swyx, you had a comment about how does this relate to sort of context window and LLMs.[00:45:37] Nikhila Ravi: And both of these types of memories have some relation to context window, so they both provide different types of information on the spatial side or in terms of the concept of the objects that we want to track. And so we found that having like six frame length for the spatial memory, Coupled with this longer period of the object pointer memory provides strong video segmentation accuracy at high speed.[00:46:01] Nikhila Ravi: So, as I mentioned, the real time aspect is really important. We have to find this speed accuracy trade off. And one way in which we sort of circumvent this is by allowing additional prompts on subsequent frames. So even if the model makes a mistake, maybe it loses the object. After an occlusion, you can provide another prompt, which actually goes into the memory.[00:46:24] Nikhila Ravi: And so the prompted frames are always in the memory. And so if you provide a prompt on a frame, we will, or the model will always remember what you provided. And so that's a way in which we can sort of avoid some of the model failure cases that actually is a big limitation of current models, current video object segmentation models.[00:46:45] Nikhila Ravi: Don't allow any way to recover if the model makes a mistake. And so, Joseph, going back to your point about the demo, that's something that we found just by playing with these models. There's no way to make a correction, and in many real world use cases, like, it's not going to be a one time prediction, but you actually want to be able to intervene, like, if an LLM makes a mistake, you can actually be like, no, actually do it this way, and provide feedback, and so, We really want to bring some of that thinking into how we build these computer vision models as well.[00:47:16] "Context Length" in Memory Attention[00:47:16] swyx: Amazing. My main reaction to finding out about the context length of eight input frames and six pass frames as their default is why not 60? Why not 600? In text language models, we're very used to severely extending context windows. And what does that do to the memory of your model?[00:47:35] Nikhila Ravi: So I think maybe one, one thing that's different is that the object in video, it is challenging.[00:47:41] Nikhila Ravi: Objects can, you know, change in appearance. There's different lighting conditions. They can deform, but I think a difference to language models is probably the amount of context that you need is significantly less than maintaining a long multi time conversation. And so, you know, coupling this. Short term spatial memory with this, like, longer term object pointers we found was enough.[00:48:03] Nikhila Ravi: So, I think that's probably one difference between vision models and LLMs.[00:48:09] Object Tracking[00:48:09] Joseph Nelson: I think so. If one wanted to be really precise with how literature refers to object re identification, object re identification is not only what SAM does for identifying that an object is similar across frames, It's also assigning a unique ID.[00:48:25] Joseph Nelson: How do you think about models keeping track of occurrences of objects in addition to seeing that the same looking thing is present in multiple places?[00:48:37] Nikhila Ravi: Yeah, it's a good question. I think, you know, SAM2 definitely isn't perfect and there's many limitations that, you know, we'd love to see. People in the community help us address, but one definitely challenging case is where there are multiple similar looking objects, especially if that's like a crowded scene with multiple similar looking objects, keeping track of the target object is a challenge.[00:49:03] Nikhila Ravi: That's still something that I don't know if we've solved perfectly, but again, the ability to provide refinement clicks. That's one way to sort of circumvent that problem. In most cases, when there's lots of similar looking objects, if you add enough refinement clicks, you can get the perfect track throughout the video.[00:49:22] Nikhila Ravi: So definitely that's one way to, to solve that problem. You know, we could have better motion estimation. We could do other things in the model to be able to disambiguate similar looking objects more effectively.[00:49:35] swyx: I'm just interested in leaving breadcrumbs for other researchers, anyone interested in this kind of architecture.[00:49:41] swyx: Like, are there papers that you would refer people to that are influential in your thinking or, you know, have, have other interesting alternative approaches?[00:49:49] Nikhila Ravi: I think there's other ways in which you can do tracking and video. You might not even need the full mask. I think that's it. Some other works that just track like points on objects.[00:49:59] Nikhila Ravi: It really, really depends on what your application is. Like if you don't care about the entire mask, you could just track a bounding box. You could just track a point on an object. And so having the high fidelity mask might not actually be necessary for certain use cases. From that perspective, you might not need the full capabilities.[00:50:19] Nikhila Ravi: of SAM or SAM2. There's many different approaches to tracking, I think I would encourage people to think about like what actually they need for their use case and then try to find something that that fits versus, yeah, maybe SAM2 is too much, you know, maybe you don't even need the full mask.[00:50:37] swyx: Makes total sense, but you have solved the problem that you set out to solve, which is no mean feat, which is something that we're still appreciating even today.[00:50:44] The Future of FAIR[00:50:44] swyx: If there are no further questions, I would just transition to sort of forward looking, future looking stuff. Joseph already hinted at, like, you know, our interest in SAM and the future of SAM, and obviously you're the best person to ask about that. I'm also interested in, like, How should external people think about FAIR, you know, like there's this stuff going on, this llama, this chameleon, this voice box, this image bind, like, how is, how are things organized?[00:51:09] swyx: And, you know, where are things trending?[00:51:11] Nikhila Ravi: Yeah, so in FAIR, we, you know, we have a number of different research areas. I work in an area called perception. So we built vision systems that solve basically, Look at all the fundamental problems in Compute Division. Can we build a step change in all of these different capabilities?[00:51:29] Nikhila Ravi: SAM was one example. SAM2 is another example. There are tons of other problems in Compute Division where we've made a lot of progress, but can we really say that they're solved? And so that's really the area in which I work on. And then there's a number of other research areas in language and in embodied AI.[00:51:49] Nikhila Ravi: And more efficient models and various other topics. So fair in general is still very much pushing the boundaries on solving these foundational problems across different domains. Well,[00:52:07] swyx: fair enough, maybe just outside of fair, just the future of computer vision, right?[00:52:10] CVPR, Trends in Vision[00:52:10] swyx: Like you are very involved in the community. What's the talk of the town at CVPR? Both of you went, who's doing the most interesting work? It's a question for both of you.[00:52:19] Joseph Nelson: I think the trends we're seeing towards more zero shot capability for common examples will accelerate. I think Mutu modality, meaning using, you know, images in tandem with text for richer understanding or images and video in tandem with audio and other mixed media will be a continued acceleration trend.[00:52:43] Joseph Nelson: The way I kind of see the field continuing to progress, the problem statement of computer vision is making sense of visual input. And I think about the world as the things that need to be observed follow your traditional bell curve, where like things that most frequently exist out in the world are on the center of that bell curve.[00:53:05] Joseph Nelson: And then there's things that are less frequently occurring that are in those long tails. For example, you know, as back as like 2014, you have the Cocoa data set, which sets out to say, Hey, can we find 80 common objects in context, like silverware and fridge and these sorts of things. And we also conceptualized the challenge of computer vision in terms of breaking it down into individual task types, because that's like the tools we had for the day.[00:53:29] Joseph Nelson: So that's why, you know, you have the origination of classification, object detection, instant segmentation. And then as you see things continue to progress. You have models and things that need to observe areas in the long tails. And so if you think of the Cocoa dataset as the center of that bell curve, I think of like the long tails, like really edge case problems.[00:53:49] Joseph Nelson: Some of our customers like Rivian, for example, only Rivian knows what the inside of like a Rivian should look like as it's assembled and put together before it makes its way to a customer and they're making custom parts. Right? So how could a model you've been trained on the things that go inside the componentry of producing a vehicle and Andreesen, What's kind of happening with computer vision is you're seeing models that generalize in the middle of the bell curve push outward faster.[00:54:17] Joseph Nelson: That's where you see the advent of like open text models or the richness of understanding of multimodal models. To allow richer understanding without perhaps any training, or maybe just using pre training and applying it to a given problem. And then, there's like, you know, kind of like the messy middle in between those two, right?[00:54:38] Joseph Nelson: So like, Akila kind of talked about examples where SAM does well out of distribution, where like, it finds an octopus, even though there wasn't octopi in the training data. I showed an example where, like, screenshots, where Sam isn't yet super great at screenshots, so maybe that's, like, in the messy middle or in the longer tails for now.[00:54:54] Joseph Nelson: But what's going to happen is there needs to be systems of validating the point of view that I think about, like, tooling to also validate that models are doing what we want them to do, adapting to datasets that we want them to adapt to. And so there's a lot of things on a forward looking basis that allow propelling that expansion of generalizability.[00:55:14] Joseph Nelson: That's for open text problems. That's where scaling up of training, of dataset curation, continues to play a massive role. Something that's notable, I think, about SAM2 is it's, what, 57, 000 videos? 51,[00:55:30] Nikhila Ravi: 000 videos? About 51, 000, yeah.[00:55:32] Joseph Nelson: And 100, 000 internal datasets. That's, like, not Massive, right? And the model size also isn't, you know, the largest, largest model being a couple hundred million parameters.[00:55:43] Joseph Nelson: The smallest model is 38 million parameters and can run at 45 FPS on an A100, right? Like the capabilities of, we're going to see more capable, more generalizable models. Being able to run on a higher wide array of problems with zero or multi shot capability on a faster, a faster rate. And I think the architecture innovations and things like SAM2 of memory, of increasingly like transformers making their way into division and probably blended architectures increasingly too.[00:56:15] Joseph Nelson: So my viewpoint of like on a go forward basis is we will have that bell curve of what humans can see both in the center of that curve and the long tails. And architectural changes allow richer understanding, multi and zero shot, and putting those into systems and putting those into industry and putting those into contexts that allow using them in practical and pragmatic ways.[00:56:38] Joseph Nelson: Nicola, I'd love to hear like your thought and perspective of like how you think the research trends map or don't map to that. And like maybe some of the key innovations that you saw at CVPR this year that, you know, Got you excited about the direction and maybe some promising early directions that you're thinking about researching or pushing the boundaries of further.[00:56:56] Nikhila Ravi: Yeah, I just wanted to actually reply to a couple of things that you said about so actually in video object segmentation, the number of classes. that are annotated in these, and then the size of these datasets are really small. So with SAM, it's, you know, we had a billion masks, we had 11 million images, didn't have class labels.[00:57:17] Nikhila Ravi: But even before that, there were a lot of datasets that have class labels and are annotated. With significantly more with, with like a lot of class labels, whereas in video datasets, the number of class labels are very small. So there's like YouTube VOS, which has 94 object categories, there's Mose, which has around like 30 or so object categories.[00:57:38] Nikhila Ravi: And they're usually like people, there's cars, there's dogs and cats and all these common objects, but not really, they don't really cover a very large number of object categories. And so while Sam learned this general notion of what an object is in an image. These video tracking models actually don't have that knowledge at all.[00:58:01] Nikhila Ravi: And so that's why having this data set is really important for the segment anything capability in video because if you just provide the mask as the input to an off the shelf Video object segmentation model. It might not actually be able to track that arbitrary object mask as effectively as a SAM2 model that's actually trained to track.[00:58:24] Nikhila Ravi: Any object across the entire video. So doing these sort of combining two models together to try to get a capability that will actually only get you so far and being able to actually create that the dataset to enable that anything capability, it was actually really important and we can actually see that when we do comparisons with baselines where we provide some two with the same input mask and the baseline model with the same input mask.[00:58:53] Nikhila Ravi: For example, the t shirt of a person, SAM2 can track the t shirt effectively across the entire video, whereas these baselines might actually start tracking the entire person, because that's what they're used to doing, and isolating it to just one part of the person is not something they were ever trained to do, and so those are sort of some of the limitations.
Automation, Alastair Method, Ask others about what they use, Annotation, ADHD, Accountability, Adaptability, Aesthetics, Agility, Ambition, Analysis, Apps, Archives, Articulation, Assessment... and AI Continue reading →
Find any powerful story in Scripture—then look closely. Past the obvious. Behind the headliners. You will likely read about a woman or man, perhaps a servant or onlooker, who contributes a line, a word, or a gesture. The servant girl who offers a word of advice to a king. A face in the crowd that sparks a ministry. Those secondary stories that represent the minor key in the symphony. Every one of them important. In this series, we will consider the unsung, unnoticed, often unnamed characters in scripture without whose contribution the music may never have been heard. The sermon today is titled "The Recommender." It is the second installment in our series "Supporting Cast: Minor Characters, Major Lessons." The Scripture reading is from 2 Kings 5:1-4 (ESV). Originally preached at the West Side Church of Christ (Searcy, AR) on July 28, 2024. All lessons fit under one of 5 broad categories: Begin, Discover, Grow, Learn, and Serve. This sermon is filed under SERVE: Making A Difference.Click here if you would like to watch the sermon or read a transcript.Sources of Inspiration for the Lesson Used in Today's Podcast:Tim Keller, “How We Live As Believers” (2012).Imaculee Ilibageza, Left To Tell. Le Miserables, Act 1.Hieronymus Weller von Molsdorf (1499-1572), “Annotations on 2 Kings.” In Weller, Liber secundus Regum, 15v. Quoted in Reformation Commentary on Scripture, OT Vol 5: 1-2 Samuel, 1-2 Kings, 1-2 Chronicles.T. R. Hobbs, 2 Kings, WBC, Vol 13.David T. Lamb, 1 & 2 Kings, The Story of God Bible Commentary.I'd love to connect with you!Watch sermons and find transcripts at nathanguy.com.Follow along each Sunday through YouTube livestream and find a study guide and even kids notes on the sermon notes page.Follow me @nathanpguy (facebook/instagram/twitter)Subscribe to my email newsletter on substack.
Topics covered in this episode: Marimo: “Future of Notebooks” pytest 8.3.0 & 8.3.1 are out Python Language Summit 2024 bash-dungeon Extras Joke Watch on YouTube About the show Sponsored by us! Support our work through: Our courses at Talk Python Training The Complete pytest Course Patreon Supporters Connect with the hosts Michael: @mkennedy@fosstodon.org Brian: @brianokken@fosstodon.org Show: @pythonbytes@fosstodon.org Join us on YouTube at pythonbytes.fm/live to be part of the audience. Usually Tuesdays at 10am PT. Older video versions available there too. Finally, if you want an artisanal, hand-crafted digest of every week of the show notes in email form? Add your name and email to our friends of the show list, we'll never share it. Michael #1: Marimo: “Future of Notebooks” via Matt Wilkie An open-source reactive notebook for Python Run one cell and marimo reacts by automatically running affected cells, eliminating the error-prone chore of managing notebook state. Marimo's reactive UI elements, like dataframe GUIs and plots, make working with data feel refreshingly fast, futuristic, and intuitive. Rapidly experiment with code and models Bind UI elements to Python values Pick-up-and-play design, with depth for power users See the FAQ Brian #2: pytest 8.3.0 & 8.3.1 are out Real excited to get --xfail-tb flag added This detaches xfail tracebacks from -rx/-ra (which was how it was pre-8.0) Keyword matching for marker expressions, that's fun. pytest -v -m "device(serial='123')" --no-fold-skipped allows for explit reporting of names of skipped tests Plus many more improvements, bug fixes, and doc improvements Michael #3: Python Language Summit 2024 Should Python adopt Calendar Versioning?: talk by Hugo van Kemenade Python's security model after the xz-utils backdoor: talk by Pablo Galindo Salgado Native Interface and Limited C API: talks by Petr Viktorin and Victor Stinner Free-threading ecosystems: talk by Daniele Parmeggiani Python on Mobile: talk by Malcolm Smith PyREPL -- New default REPL written in Python: talk by Pablo Galindo Salgado, Łukasz Langa, and Lysandros Nikolaou Should we make pdb better?: talk by Tian Gao Limiting yield in async generators: talk by Zac Hatfield-Dodds Annotations as Transforms: talk by Jason R. Coombs Lightning Talks, featuring talks by Petr Viktorin, David Hewitt, Emily Morehouse, Łukasz Langa, Pablo Galindo Salgado, and Yury Selivanov Brian #4: bash-dungeon “This game is intended to teach new users how to use their shell in a fun and interactive way.” Just clone the repo and start exploring with cd, ls, and cat. First moves cd bash-dungeon ls cd Enter ls cat parchment A fun way to learn some commands you might need and/or might have forgotten about. Extras Brian: Python 3.12.0b4, final beta, is out If hanging out on discuss.python.org, please checkout Community Guidelines And if it's still not clear why we need these, check out Inclusive communications expectations in Python spaces Google Chrome news Michael: PySimpleGUI goes commercial with obfuscated “source open”? Still have seats for Code in a Castle event Reactive Dashboards with Shiny for Python free course Joke: 40 Million in in Series A Funding - may be a lot of reading, but I found it funny Thanks to VM Brasseur for sharing this one. Also a few from pyjokes 0.7.2 (first new version since 2019) If at first you don't succeed, call it version 1.0. A product manager walks into a bar, asks for drink. Bartender says no, but will consider adding later. Triumphantly, Beth removed Python 2.7 from her server in 2030. 'Finally!' she said with glee, only to see the announcement for Python 4.4.1 Although, if CalVer, PEP 2026, happens, that'll just be Python 3.30.0.
In this episode, Ryan is joined by Jeff Mills, a veteran in the AI and machine learning space. With over 25 years of experience, Jeff shares invaluable insights on scaling businesses, particularly in the AI industry. From his early days at Yahoo to his current role as President at iMerit, Jeff offers a unique perspective on growing companies while maintaining a social impact focus. Join 2,500+ readers getting weekly practical guidance to scale themselves and their companies using Artificial Intelligence and Revenue Cheat Codes. Explore becoming Superhuman here: https://superhumanrevenue.beehiiv.com/ KEY TAKEAWAYS iMerit works with leading AI companies in mobility, tech, and medical verticals, providing data solutions and annotation services for AI model development. Jeff talks about the importance of high-quality data in AI development, comparing it to a chef selecting the best ingredients for a Michelin-star restaurant. iMerit has evolved to include automation and a platform for data labelling, while still maintaining human-in-the-loop processes for validation and verification. The company employs a workforce pyramid, ranging from general workers to highly specialized experts in various fields, to meet diverse AI development needs. Retrieval Augmentation Generation (RAG) models are expected to be a significant trend in AI development over the next year and a half. iMerit adapts to customer needs, focusing on problem-solving and developing expertise in specific domains as required by clients. Jeff stresses the importance of domain expertise in prompt engineering and evaluating AI model outputs. BEST MOMENTS "You are literally a prompt engineer. You have created the box." "We like to write, but we also are good readers. We want to lead by example, but we also are good at reading, hearing what someone's challenges are, what their problems are. And ultimately we're in the problem solving business.” "We've built a workforce pyramid, so there's certain work that can be done if you have kind of expertise at a wide level." "I've been in AI for about, you know, 25 years in different ways." Ryan Staley Founder and CEO Whale Boss ryan@whalesellingsystem.com www.ryanstaley.io Saas, Saas growth, Scale, Business Growth, B2b Saas, Saas Sales, Enterprise Saas, Business growth strategy, founder, ceo: https://www.whalesellingsystem.com/closingsecrets
How can social annotation transform traditional reading into a collaborative learning experience?Damijana Keržič and Vida Zorko from the University of Ljubljana delve into this question through their research on Diigo, a social annotation tool. They investigate its impact on student motivation, comprehension, and the correlation between learning approaches and online activity.Read the original research: doi.org/10.1080/2331186X.2023.2269043
Unveiling Accelerated Business Success by Unlocking AI Potential We welcome Joe Buggy to this week's episode of the Digitally Irresistible podcast. As an innovative executive leader with a rich background in operations, business development, and finance, with specialization in the BPO sector, Joe is renowned for his strategic insights. Growing up as the son of an Air Force family with Irish-Italian heritage, Joe developed a keen eye for detail and a knack for problem solving. His passion for optimizing processes and delivering results, fueled by his experiences working alongside industry-leading professionals, has shaped his career trajectory. Leveraging his deep expertise in trust and safety and content management, Joe has led the charge on multiple transformative endeavors for business process outsourcing (BPO) companies, propelling growth and performance within these customer-centric enterprises. In this episode, we delve into the world of data annotation and labeling and its impact on the business world. Exploring Content Management and Data Annotation To provide context, we first explore the realm of content management—a cornerstone of brand representation and engagement in the digital age. Joe explains how content management encompasses everything from digital presence to product portrayal, emphasizing its pivotal role in shaping brand perception and customer experience. Transitioning to the core of our discussion, Joe breaks down the concepts of data annotation and labeling, which are critical aspects of content management since they ensure a brand's content is accurately described in its systems. He explains that labeling involves assigning simple tags to unstructured data, such as images or text, to facilitate understanding of artificial intelligence (AI) algorithms. Joe gives an example of a cat image, where the label "cat" informs the system about the content, demonstrating that this process extends to all forms of data. Annotation, however, adds layers of context, enabling more nuanced interpretation and data utilization for sentiments, uses, or directions. If we consider four primary data types—numerical/alphanumeric text, images, audio, and video—the complexity and unstructured nature increase as we move from numeric to alphanumeric to image, audio, and video data. This escalation underscores the crucial need for labeling and annotation to provide context for AI models. For example, in image recognition, labeling each image with metadata such as "flower species" enables AI to accurately classify different types of flowers. Similarly, in audio transcription, labeling with timestamps and the speaker identities ensures precise transcription of conversations. In video analysis, annotations like "suspicious behavior" help AI detect and respond to specific events. Overall, labeling and annotation are essential for transforming raw data into structured information that AI can effectively understand and utilize across various applications. The Intersection of Annotation, Industry Applications, and Deliberate Partnerships in AI Development In our deep dive into the realm of AI development, Joe further illuminates the pivotal role of annotation and labeling. He explains how these foundational processes serve as the bedrock for training AI models, elevating their accuracy and contextual understanding to unprecedented levels. Joe underscores the importance of structured data in this process, emphasizing how it enables AI algorithms to glean meaningful insights and make more accurate predictions that drive successful outcomes for brands. As we cross the landscape of data annotation and labeling, Joe provides a panoramic view of their diverse applications spanning numerous sectors. From the dynamic realms of health care, where AI powers telemedicine and aids in drug development, to the bustling domains of retail , where every retailer strives for a seamless omnichannel customer experience (CX) Joe explains how AI-driven solutions create transformative changes. In health care, AI models assist in diagnosing medical problems and understanding drug interactions by relying on meticulously labeled data. Similarly, in retail, AI improves customer experiences by allowing users to virtually try on clothing or eyeglasses tailored to their body style or face shape. These algorithms continuously learn from user preferences, suggesting products that align with individual tastes, akin to the automotive industry's use of AI for autonomous vehicles and predictive maintenance. Across digitally native industries, travel services, consumer products, and gaming, AI's integration optimizes operations, predicts market trends, and fosters brand acceptance through data-driven insights and personalized recommendations. Given the scale and complexity inherent in data annotation, Joe describes the importance of forging partnerships with BPO organizations. Joe highlights how these collaborations empower brands to navigate the intricate landscape of AI development with confidence and agility. By tapping into BPOs' depth of knowledge in annotating and labeling data—whether through bounding box, semantic annotation, video annotation , or cuboids—brands can ensure high-quality data preparation crucial for computer vision, natural language processing, and audio processing applications. BPOs excel by identifying and hiring top talent and training them rigorously in specialized systems and processes. Moreover, these partnerships enable continuous improvement through robust quality monitoring, feedback mechanisms, and coaching to drive new goals and introduce optimized processes. Through strategic collaborations, Joe envisions a future where innovation knows no bounds and the transformative potential of AI is fully unleashed to shape a brighter tomorrow. With support from BPOs, organizations can confidently build and execute their AI strategies with the scalability, quality, and security needed for success. Navigating Security, Privacy, and Brand Considerations in AI Initiatives In our exploration of AI initiatives, Joe delves into the critical aspects of data security and privacy. Addressing pertinent concerns surrounding the handling of consumer and proprietary data, Joe emphasizes the need for robust measures to safeguard sensitive information and the importance of implementing stringent protocols and cutting-edge technologies to ensure compliance with regulatory standards and instill trust among stakeholders. By prioritizing security and privacy in AI-driven initiatives, organizations can mitigate risks and uphold the integrity of their data assets, paving the way for sustainable growth and innovation in the digital landscape . With significant experience in navigating the complexities of AI implementation, Joe's valuable insights highlight key considerations that can shape the success of brands seeking to harness the full potential of AI. He points out the significance of aligning AI strategies with organizational goals and values , ensuring a cohesive approach toward driving business objectives. Identifying gaps in expertise and resources and forming tactical partnerships with trusted providers can help augment a company's capabilities and ensure seamless execution of services. Adopting a holistic approach and leveraging the expertise of external partners enables brands to unlock the full potential of AI technology, driving innovation and sustainable business growth in today's competitive landscape. "Identify where [your brand's] gaps are and if those gaps include meeting the speed, the scale, the different data types, and the security at a level of accuracy and consistency that the organization requires, I would look to partner with a trustworthy organization to address those gaps." - Joe Buggy What Joe Likes to Do for Fun When not working, Joe enjoys outdoor cooking and golf, highlighting the importance of work-life balance and sharing cherished moments with friends and family. To learn more about Joe, connect with him on LinkedIn. Watch the video here. Read the blog post here.
Learn how Andrea Brown, CEO of Reliabl, is revolutionizing AI by ensuring diverse communities are represented in data annotation. Discover how this approach not only reduces bias but also improves algorithmic performance. Andrea shares insights from her journey as an entrepreneur and AI researcher. Episode timestamps(02:22) Andrea's Career Journey and Experience with Open Source (Adobe, Macromedia, and Alteryx)(11:59) Origins of Alteryx's AI and ML Capabilities / Challenges of Data Annotation and Bias in AI(19:00) Data Transparency & Agency(26:05) Ethical Data Practices(31:00) Open Source Inclusion Algorithms(38:20) Translating AI Governance Policies into Technical Controls(39:00) Future Outlook for AI and ML(42:34) Impact of Diversity Data and Inclusion in Open SourceQuotesAndrea Brown"If we get more of this with data transparency, if we're able to include more inputs from marginalized communities into open source data sets, into open source algorithms, then these smaller platforms that maybe can't pay for a custom algorithm can use an algorithm without having to sacrifice inclusion." Charna Parkey“I think if we lift every single platform up, then we'll advance all of the state of the art and I'm excited for that to happen."Connect with AndreaConnect with Charna
Dans ce long…. épisode, Emmanuel, Guillaume et Arnaud discutent de l'actualité avec Chicori (un runtime WASM en Java), Jakarta Data, Quarkus 3.10, Spring AI, Hibernate 6.5, mais aussi quelques retours aux basiques (timezones, rate limiting, …). Gros focus sur les nouveautés annoncées à Google I/O 2024 et dans l'écosystème IA en général avec les annonces d'OpenAI, Claude, Grok et d'autres. Différents outils sont aussi couverts comme Git, IntelliJ, ASDF, BLD, S3. Et enfin des sujets sur la haute disponibilité de Keycloak, la ré-indexation sans downtime, les challenges des implémentations alternatives, le mode vigilant dans GitHub, Redis et les changements de license, et les investissements de Microsoft et AWS en France dans le cadre du programme #ChooseFrance. N'hésitez pas à nous soumettre vos questions sur https://lescastcodeurs.com/ama nous y répondrons dans les prochains épisodes. Enregistré le 17 mai 2024 Téléchargement de l'épisode LesCastCodeurs-Episode-312.mp3 News Langages Un runtime WASM en Java https://github.com/dylibso/chicory Projet tout nouveau, encore loin de la maturité Mais intéressant à suivre pour exécuter du code WebAssembly dans une application Java le projet n'a pas 15 jours non plus quand même :) Faire tourner des plugins WASM dans la JVM (e.g. plugins) On peut faire des heap dump en cas de OutOfMemoryException en compilation native https://quarkus.io/blog/heapdump-oome-native/ depuis JDK 21 Un exemple avec Quarkus Et le GC epsilon 100 exercices pour se mettre à Rust https://rust-exercises.com/ Librairies Hibernate 6.5 est sorti https://in.relation.to/2024/04/25/orm-650/ cache full pour les entités et leur collections (le défaut est shallow) Java record pour les @IdClass Les filtres peuvent être auto activés par défaut (vs à faire sur chaque session). Les filtres sont pas mal pour gérer par exemple des soft delete Keybased pagination pour éviter les trous de résultant en cas de modification d'entités en parallèle de.une recherche paginée. S.appuie sur une clé unique et ordonnée genre ISBN Une tech preview de Jakarta Data En parlant de Jakarta Data, deux articles sur le sujet https://in.relation.to/2024/04/01/jakarta-data-1/ https://in.relation.to/2024/04/18/jakarta-data-1/ concept de repository pas lié à une entité mais à une relation logique entre les recherches interagit via stateless session et est un bean CDI Code généré bien sur 4 opérateur crud et les requêtes save est up sert Type sage au sens ou le nom des méthodes n'est pas la logique de recherche Annotation et nom des paramètres et c'est type safe via un annotation processor ou string dans @Query qui est type safe aussi via le processeur discute plus de type safety et pagination Quarkus 3.10 avec quelques nouveautés https://quarkus.io/blog/quarkus-3-10-0-released/ flyway 10 arrive avec support natif Hibernate search supporte le standalone POJO mapper notamment pour elastic search (pas que ORM) Modification des propriétés Quarkus.package automatiquement remplacées par quarkus update et Quarkus 3.9 a fait son grand renommage réactif https://quarkus.io/blog/quarkus-3-9-1-released/ Clarifier que les extensions réactive n'imposent pas des apis réactives et seulement leur cœur implémenté en réactif ou offre optionellement des apis reacrive Les gens pensaient à tors que les réactives imposaient le modèle de programmation la encore quarkus update à la rescousse Un article sur l'api structured output pour Spring AI https://spring.io/blog/2024/05/09/spring-ai-structured-output un article descriptif sur quand cette api est utilisée Et les détails de son usage Comment passer une TimeZone dans spring boot et ce que cela impacte en terme de composants https://www.baeldung.com/spring-boot-set-default-timezone du basique mais toujours utile Task ou app Programmatiquement Sur certains lifecycles de Spring Infrastructure Un article et la vidéo de Devoxx France sur la haute disponibilité de Keycloak, comment c'est implémenté https://www.keycloak.org/2024/05/keycloak-at-devoxx-france-2024-recap l'infra d'identité est une infra clé Donc gérer la haute disponibilité est critique C'est un article qui pointe sur une vidéo de Devoxx France et la doc de keycloak sur comment tout cela est implémenté Cloud Comment se ruiner avec des buckets S3 https://medium.com/@maciej.pocwierz/how-an-empty-s3-bucket-can-make-your-aws-bill-explode-934a383cb8b1 Amazon fait payer pour les requêtes non autorisées Il suffit de connaître le nom d'un bucket pour faire payer son propriétaire Amazon travaille pour fournir une solution / un fix. il est tombé par hasard sur un nom de bucket utilisé « pour de faux » par un outil open source populaire Bien rajouter un suffixe à ses buckets peut réduire le risque Mais pas l'éliminer un fix a été livré par amazon https://aws.amazon.com/about-aws/whats-new/2024/05/amazon-s3-no-charge-http-error-codes/ Data et Intelligence Artificielle Guillaume résume GoogleIO https://x.com/techcrunch/status/1790504691945898300?s=61&t=WImtt07yTQMhhoNPN6lYEw AI overview plus besoin d'aller sur les sites Google I/O 2024 Google I/O 2024 résumé en vidéo de 10 minutes https://www.youtube.com/watch?v=WsEQjeZoEng et en 100 bullet points https://blog.google/technology/ai/google-io-2024-100-announcements/ Message de Sundar Pichai https://blog.google/inside-google/message-ceo/google-io-2024-keynote-sundar-pichai/#creating-the-future Project Astra, un assistant universel, sur smartphone avec qui on peut avoir une conversation normale et à qui montrer avec la caméra ce qui nous entoure https://www.theverge.com/2024/5/14/24156296/google-ai-gemini-astra-assistant-live-io Nouveau modèle Gemini 1.5 Flash, quasi aussi performant que le nouveau Gemini 1.5 Pro, mais beaucoup plus rapide (premiers tokens dans la seconde) et aussi moins cher https://blog.google/technology/developers/gemini-gemma-developer-updates-may-2024/ Gemini 1.5 Pro est Gemini 1.5 Flash sont disponibles avec une fenêtre de contexte d'un million de tokens, mais il y a une liste d'attente pour tester une fenêtre de 2 millions de tokens https://aistudio.google.com/app/waitlist/97595554 https://cloud.google.com/earlyaccess/cloud-ai?e=48754805&hl=en PaliGemma un nouveau modèle de vision ouvert dans la famille Gemma (pour faire du Q&A du sous-titrage) et preview de Gemma 2, avec une version à 27 milliards de paramètres https://developers.googleblog.com/en/gemma-family-and-toolkit-expansion-io-2024/ Gemini disponible dans les IDEs : Android Studio, IDX, Firebase, Colab, VSCode, Cloud and Intellj Gemini AI Studio enfin disponible en Europe Gemini supporte le parallel function calling et l'extraction de frame dans les vidéos Trillium, la 6ème version des TPU (Tensor Processing Unit), les processeurs spécifiques ML dans Google Cloud, 5 fois plus puissant que la génération précédente et 67% plus efficace en énergie https://cloud.google.com/blog/products/compute/introducing-trillium-6th-gen-tpus Le projet NotebookLM rajoute une fonctionnalité de Audio Overview qui permet de discuter avec son corpus de documents avec une conversation vocale https://notebooklm.google.com/ On peut appliquer le “grounding” avec Google Search pour l'API Gemini, pour que le modèle Gemini puisse chercher des informations complémentaires dans Google Search https://cloud.google.com/blog/products/ai-machine-learning/vertex-ai-io-announcements Annonce de Imagen 3, la future version de du modèle de génération d'images Imagen qui améliore la qualité et possède un très bon support du texte dans les images (objectif de disponibilité à l'été) https://blog.google/technology/ai/google-generative-ai-veo-imagen-3/#Imagen-3 https://deepmind.google/technologies/imagen-3/ DeepMind annonce Veo, un nouveau modèle de génération de vidéo très convaincant qui peut faire des vidéos en 1080p de 60s, mais en combinant plusieurs prompts successifs, il peut générer des vidéos plus longues qui s'enchainent https://deepmind.google/technologies/veo/ VideoFX, ImageFX et MusicFX, des expérimentations de Google AI intégrant Imagen 3 et Veo (pas encore disponibles en Europe) https://blog.google/technology/ai/google-labs-video-fx-generative-ai/ Gemini Advanced https://blog.google/products/gemini/google-gemini-update-may-2024/#context-window Les utilisateurs de Gemini Advanced (l'application web) utilisent Gemini 1.5 Pro avec la fenêtre de contexte de 1 million de tokens, la possibilité de charger des documents de Google Drive, et bientôt la possibilité de générer des graphiques. Gemini Advanced rajoute aussi la capacité de générer des itinéraires de voyage (avec intégration de Google Flights, etc) Fonctionnalité Gemini Live pour avoir une conversation vocale naturelle avec Gemini https://blog.google/products/gemini/google-gemini-update-may-2024/#gemini-live Gem : des plugins pour Gemini Advanced pour créer ses propres assistants personnalisés https://blog.google/products/gemini/google-gemini-update-may-2024/#personalize-gems Ask Photos, on peut poser à Google Photos des questions plus complexes comme “quelle est ma plaque d'immatriculation” et Photos devine que parmi toutes les photos de voitures lequelle est certainement la nôtre et extrait le numéro de plaque https://blog.google/products/photos/ask-photos-google-io-2024/ Même dans Google Messages vous pourrez échanger avec Gemini Google Search https://blog.google/products/search/generative-ai-google-search-may-2024/ Rajout d'un modèle Gemini spécial search intégré qui permet à Google Search de répondre aux questions de la barre de recherche avec une raisonnement multi-étapes, en étant capable de faire de la planification, en mode multimodal (texte, image, vidéo, audio) Planning de repas et de voyage, supporté dans Gemini, va arriver aussi dans Search Gemini 1.5 Pro est disponible dans le panneau latéral de Gmail, Docs, Sheets, Drive https://blog.google/products/workspace/google-gemini-workspace-may-2024-updates/ SynthID va même fonctionner pour du texte https://deepmind.google/discover/blog/watermarking-ai-generated-text-and-video-with-synthid/ Gemini Nano bientôt disponible dans les prochaines version de Chrome, pour utiliser le LLM directement dans le navigateur Android Seconde béta d'Android 15 https://android-developers.googleblog.com/2024/05/the-second-beta-of-android-15.html Private space pour garder des apps secures avec un niveau d'authentification supplémentaire Google collabore avec Samsung et Qualcomm sur la réalité augmentée dans Android https://developers.googleblog.com/en/google-ar-at-io-2024-new-geospatial-ar-features-and-more/ Project Gameface arrive sur Android (pour diriger Android avec les yeux, avec les expressions du visage, pour l'accessibilité) https://developers.googleblog.com/en/project-gameface-launches-on-android/ Gemini Nano va passer en multimodal, pas juste du texte Circle to search étendu à 100 millions de téléphones supplémentaires supportant Nano et va permettre de poser des questions, par exemple pour l'aide aux devoirs des enfants https://blog.google/products/android/google-ai-android-update-io-2024/#circle-to-search Detect phone scam on device with Gemini Nano Talkback, l'application pour l'accessibilité dans Android, va tirer parti de la multimodalité de Gemini Nano Bientôt de la génération d'image qu'on pourra intégrer dans ses mails, ses messages Wear OS https://android-developers.googleblog.com/2024/05/whats-new-in-wear-os-io-24.html Travail sur l'économie d'énergie pour faire durer les montres plus longtemps avant la prochaine recharge. Par exemple, 20% de consommation en moins lorsqu'on court un marathon ! Plus de type de données pour les activités physiques Project IDX accessible sans liste d'attente https://developers.googleblog.com/en/start-building-with-project-idx-today/ Firebase annonce 3 nouveaux produits https://developers.googleblog.com/en/whats-new-in-firebase-io-24/ Data Connect, un backend-as-a-service avec PostgreSQL https://firebase.google.com/products/data-connect App Hosting, hosting d'application Next et Angular https://firebase.google.com/products/app-hosting Genkit, a GenAI framework for app developers https://firebase.google.com/products/genkit Dart 3.4 avec support de Wasm comme target de compilation https://medium.com/dartlang/dart-3-4-bd8d23b4462a OpenAI lance son nouveau modèle: gpt-4o http://openai.com/index/hello-gpt-4o/ https://x.com/openaidevs/status/1790083108831899854?s=46&t=GLj1NFxZoCFCjw2oYpiJpw Audio, vision et reconnaissance de texte en realtime Plus rapide et 50% moins cher que son prédécesseur 4-turbo https://claude.ai/ est disponible en europe Claude, le modèle est créé par Anthropic: Claude est un assistant IA basé sur un grand modèle de langage entraîné selon des principes éthiques stricts. Il accorde une grande importance à l'honnêteté, l'impartialité et le respect de l'être humain. Son raisonnement repose sur une compréhension profonde des concepts plutôt que sur de simples associations statistiques. Il cherche activement à corriger les éventuels biais ou erreurs. Claude est polyvalent et peut s'adapter à différents styles de communication et niveaux de complexité selon le contexte. Il maîtrise de nombreux domaines académiques et scientifiques. Il est capable d'introspection sur ses propres processus de pensée et ses limitations. La vie privée et la confidentialité sont des priorités pour lui. Claude continue d'apprendre et de s'améliorer grâce aux interactions avec les humains. Son but est d'être un assistant fiable, éthique et bienveillant. quelqu'un sait comment ils font pour raisonner et pas juste LLM statistiquer? Comment ils prouvent cela ? C'est du code à part? Grok le modèle de X/Twitter/Musk est aussi dispo en Europe https://x.com/x/status/1790917272355172401?s=46&t=GLj1NFxZoCFCjw2oYpiJpw un truc unique c'est qu'il utilise les tweet comme reference sur ce qu'il dit. Par exemple demande les meilleurs Java Champions et c'est sur les tweet recents , probablement une sorte de RAG ou une sorte de fine tuning sur les derniers tweets, je ne sais pas L'algorithm des modeles de diffusion expliqués https://x.com/emmanuelbernard/status/1787565568020619650 deux articles, un general et lisible l'autre plus abscon mais avec certains details interessants sur le downsizing étapes ajout de bruit à des images (learning) pour après appliquer le process opposé le reverse diffusion process On prédit le bruit à enlever, on l'enlève et on repère le processus. Et tout cela est influencé par le prompt. Reindexation sans downtime des données de documentation de Quarkus, en quarkus bien sûr https://quarkus.io/blog/search-indexing-rollover/ utilise hibernate search Utilisé Elasticsearch / opensearch Article qui explique une des approches pour reindexer sans downtime via index alias Outillage Un article qui parle de l'outil de build bld, peu connu, qui permet d'écrire ses builds simplement dans une classe Java https://sombriks.com/blog/0070-build-with-bld-and-why-it-matters/ IntelliJ 2024.1 est sorti https://blog.jetbrains.com/idea/2024/05/what-s-new-in-intellij-idea-ultimate-2024-1/ complétion de ligne entière (deep learning) Assistant AI amélioré Spring Boot support amélioré sur bean completion et génération de diagramme Support de dev containers simplifié Amélioration support quarkus avec notamment icône dev ui et config des tests Support OpenRewrite Server wiremock et plein d'autres choses En version beta public, Homebrew permet de vérifier la provenance des packages (bottles) https://blog.trailofbits.com/2024/05/14/a-peek-into-build-provenance-for-homebrew/ Basé sur le système “build provenance” de sigstore https://docs.sigstore.dev/verifying/attestation/#validate-in-toto-attestations qui repose sur les attestations in-toto https://in-toto.io/ Mettez à jour git en version 2.45.1 pour fixer des failles de sécurité https://github.blog/2024-05-14-securing-git-addressing-5-new-vulnerabilities/ CVE-2024-32002 (Critique, Windows & macOS) : Les repos Git avec des sous-modules peuvent tromper Git pour lui faire exécuter un hook (élément de script) à partir du répertoire .git/ pendant une opération de clonage, permettant l'exécution de code à distance (Remote Code Execution). CVE-2024-32004 (Important, machines multi-utilisateurs) : Un attaquant peut concevoir un repo local qui exécute du code arbitraire lors du clonage. CVE-2024-32465 (Important, toutes les configurations) : Le clonage à partir de fichiers .zip contenant des repos Git peut contourner les protections, et potentiellement exécuter des hooks malveillants. CVE-2024-32020 (Faible, machines multi-utilisateurs) : Les clones locaux sur le même disque peuvent permettre à des utilisateurs non approuvés de modifier des fichiers liés physiquement (hard link) dans la base de données des objets du repo cloné. CVE-2024-32021 (Faible, machines multi-utilisateurs) : Le clonage d'un repo local avec des liens symboliques (symlinks) peut entraîner la création de liens physiques vers des fichiers arbitraires dans le répertoire objects/. Architecture Visualisation des algorithmes de rate limitation https://smudge.ai/blog/ratelimit-algorithms Méthodologies Le problème de l'implémentation alternative https://pointersgonewild.com/2024/04/20/the-alternative-implementation-problem/ Article par un développeur qui a développé des Just-in-Time compiler pour différents langages Remarqué que développer une implémentation alternative d'un langage (par exemple) n'a jamais vraiment rencontré le succès Les gens préfèrent l'original à une alternative qui est dépendante de / a peine à suivre l'implémentation d'origine Pour son cas, sur le JIT, il a travaillé sur un JIT intégré directement dans CRuby (plutôt que faire son implémentation alternative comme TruffleRuby), et sont JIT est intégré maintenant dedans directement Plus facile de rejoindre / s'intégrer au projet plutôt que d'être une alternative pour laquelle il faut convaincre les gens de l'adopter Le mode vigilant dans GitHub https://x.com/emmanuelbernard/status/1790026210619068435 c'est la suite du blog wsur la signature des commits que j'ai fait ul y a quelques temps https://emmanuelbernard.com/blog/2023/11/27/git-signing-ssh/ Maintenant, GitHub rajoute de plus en plus d'infos si les signatures ne matchent pas ou ne sont pas présentes Loi, société et organisation Une perspective sur Redis et les changements de license par un devrel AWS OpenSearch https://www.infoworld.com/article/3715247/the-end-of-vendor-backed-open-source.html les sociétés regardent l'impact légal des licenses source available pour elles même en usage interne Ça casse l'écosystème de spécialisations au dessus du produit (logz.io au dessus d'elastic démarré avant le changement de license) Redis top 10 contribs à AWS et Alibaba er Huawei et 3 redis. Donc c'est pas redis qui contribue tout. La plupart des ingénieurs de redislab ne bossent pas sur redis OSS, mais sur cloud et entreprise Peut être la fin des single vendor oss Il n'y a que les cloud providers qui peuvent fournir du OSS sans affecter leur structure du coût C'est un ex AWS en fait. Maintenant indépendant Microsoft va investir 4 milliards en France (datacenters et IA) https://news.microsoft.com/fr-fr/2024/05/13/microsoft-announces-the-largest-investment-to-date-in-france-to-accelerate-the-adoption-of-ai-skilling-and-innovation/ Il ne sont pas les seuls dans le cadre du programme #chooseFrance https://www.info.gouv.fr/actualite/choose-france-un-record-de-15-milliards-deuros-dinvestissements-etrangers Mais cela n'est pas sans laisser de questions sur l'avenir de notre activité avec les US qui externalisent désormais leur silicon valley https://www.cybernetica.fr/la-france-laboratoire-de-la-silicon-valley-2-0/ Outils de l'épisode ASDF un gestionnaire de version multi-runtime https://asdf-vm.com Arnaud l'avait recommandé mais je restais sur rvm apres des deboires, je suis passé a asdf, qui fonctionne mais pour le jdk j'utilise sdkman pour les javaistes ca parrait plus poussé Conférences Les videos de Devoxx France sont en ligne https://www.youtube.com/playlist?list=PLTbQvx84FrARars1vXos7mlPdvYJmsEoK La liste des conférences provenant de Developers Conferences Agenda/List par Aurélie Vache et contributeurs : 16-17 mai 2024 : Newcrafts Paris - Paris (France) 22 mai 2024 : OpenInfra Day France - Palaiseau (France) 22-25 mai 2024 : Viva Tech - Paris (France) 24 mai 2024 : AFUP Day Nancy - Nancy (France) 24 mai 2024 : AFUP Day Poitiers - Poitiers (France) 24 mai 2024 : AFUP Day Lille - Lille (France) 24 mai 2024 : AFUP Day Lyon - Lyon (France) 28-29 mai 2024 : Symfony Live Paris - Paris (France) 1 juin 2024 : PolyCloud - Montpellier (France) 6 juin 2024 : WAX 2024 - Aix-en-Provence (France) 6-7 juin 2024 : DevFest Lille - Lille (France) 6-7 juin 2024 : Alpes Craft - Grenoble (France) 7 juin 2024 : Fork it! Community - Rouen (France) 11 juin 2024 : Cloud Toulouse - Toulouse (France) 11-12 juin 2024 : OW2con - Paris (France) 11-12 juin 2024 : PGDay Lille - Lille (France) 12-14 juin 2024 : Rencontres R - Vannes (France) 13-14 juin 2024 : Agile Tour Toulouse - Toulouse (France) 14 juin 2024 : DevQuest - Niort (France) 18 juin 2024 : Mobilis In Mobile 2024 - Nantes (France) 18 juin 2024 : BSides Strasbourg 2024 - Strasbourg (France) 18 juin 2024 : Tech & Wine 2024 - Lyon (France) 19-20 juin 2024 : AI_dev: Open Source GenAI & ML Summit Europe - Paris (France) 19-21 juin 2024 : Devoxx Poland - Krakow (Poland) 26-28 juin 2024 : Breizhcamp 2024 - Rennes (France) 27 juin 2024 : DotJS - Paris (France) 27-28 juin 2024 : Agi Lille - Lille (France) 4-5 juillet 2024 : Sunny Tech - Montpellier (France) 8-10 juillet 2024 : Riviera DEV - Sophia Antipolis (France) 6 septembre 2024 : JUG Summer Camp - La Rochelle (France) 6-7 septembre 2024 : Agile Pays Basque - Bidart (France) 17 septembre 2024 : We Love Speed - Nantes (France) 19-20 septembre 2024 : API Platform Conference - Lille (France) & Online 25-26 septembre 2024 : PyData Paris - Paris (France) 26 septembre 2024 : Agile Tour Sophia-Antipolis 2024 - Biot (France) 2-4 octobre 2024 : Devoxx Morocco - Marrakech (Morocco) 7-11 octobre 2024 : Devoxx Belgium - Antwerp (Belgium) 10 octobre 2024 : Cloud Nord - Lille (France) 10-11 octobre 2024 : Volcamp - Clermont-Ferrand (France) 10-11 octobre 2024 : Forum PHP - Marne-la-Vallée (France) 11-12 octobre 2024 : SecSea2k24 - La Ciotat (France) 16 octobre 2024 : DotPy - Paris (France) 17-18 octobre 2024 : DevFest Nantes - Nantes (France) 17-18 octobre 2024 : DotAI - Paris (France) 30-31 octobre 2024 : Agile Tour Nantais 2024 - Nantes (France) 30-31 octobre 2024 : Agile Tour Bordeaux 2024 - Bordeaux (France) 31 octobre 2024-3 novembre 2024 : PyCon.FR - Strasbourg (France) 6 novembre 2024 : Master Dev De France - Paris (France) 7 novembre 2024 : DevFest Toulouse - Toulouse (France) 8 novembre 2024 : BDX I/O - Bordeaux (France) 13-14 novembre 2024 : Agile Tour Rennes 2024 - Rennes (France) 21 novembre 2024 : DevFest Strasbourg - Strasbourg (France) 28 novembre 2024 : Who Run The Tech ? - Rennes (France) 3-5 décembre 2024 : APIdays Paris - Paris (France) 4-5 décembre 2024 : Open Source Experience - Paris (France) 22-25 janvier 2025 : SnowCamp 2025 - Grenoble (France) 16-18 avril 2025 : Devoxx France - Paris (France) Nous contacter Pour réagir à cet épisode, venez discuter sur le groupe Google https://groups.google.com/group/lescastcodeurs Contactez-nous via twitter https://twitter.com/lescastcodeurs Faire un crowdcast ou une crowdquestion Soutenez Les Cast Codeurs sur Patreon https://www.patreon.com/LesCastCodeurs Tous les épisodes et toutes les infos sur https://lescastcodeurs.com/
An airhacks.fm conversation with Anton Arhipov (@antonarhipov) about: Anton appeared previously on "#273 The Long Road to Java and Kotlin", discussion about Anton Arhipov's artwork using circles and a compass, attending the JVM Language Summit in 2011 where Kotlin was introduced by JetBrains, initial skepticism about the need for a new JVM language, JSR-305 Annotations for Software Defect Detection by William Pugh, Kotlin's null safety features and interoperability with Java, Kotlin's growth and adoption by Android developers, Kotlin's multiplatform capabilities for targeting native, JavaScript, and WebAssembly, Kotlin's potential beyond Android development, Kotlin's core libraries for date/time, serialization, and coroutines, the Kotlin compiler being self-hosted and written in Kotlin, benefits of Kotlin Native for serverless and IoT compared to GraalVM, Kotlin Multiplatform support in the upcoming JetBrains Fleet IDE, designers using similar UI principles across IDEs and applications Anton Arhipov on twitter: @antonarhipov
Annotation can be a powerful way to improve comprehension and increase engagement, but its effectiveness can vary depending on how it's taught. In this episode, two teachers share their classroom-tested approaches to teaching students how to effectively annotate texts: 3rd grade teacher Andrea Castellano and high school English teacher Irene Yannascoli. Thanks to Listenwise and Studyo for sponsoring this episode. To read a full transcript of this conversation, visit cultofpedagogy.com/art-of-annotation/.
I'm not sure if I've ever loved recording an episode more than I loved this one. I am CRYING!!!!!!!!!!
In today's flashback, an outtake from Episode 762, my conversation with author John Keene about his poetry collection Punks, which won the National Book Award for Poetry in 2022. The episode first aired on March 9, 2022. Keene is a writer, translator, professor, and artist who was named a MacArthur Fellow in 2018. In 1989, Keene joined the Dark Room Writers Collective, and is a Graduate Fellow of the Cave Canem Writers Workshops. He is the author of Annotations, and Counternarratives, both published by New Directions, as well as several other works, including the poetry collection Seismosis, with artist Christopher Stackhouse, and a translation of Brazilian author Hilda Hilst's novel Letters from a Seducer. Keene is the recipient of many awards and fellowships--including the Windham-Campbell Prize, the Whiting Foundation Prize, the Republic of Consciousness Prize, and the American Book Award. He teaches at Rutgers University-Newark. *** Otherppl with Brad Listi is a weekly literary podcast featuring in-depth interviews with today's leading writers. Available where podcasts are available: Apple Podcasts, Spotify, YouTube, etc. Subscribe to Brad Listi's email newsletter. Support the show on Patreon Merch @otherppl Instagram TikTok Email the show: letters [at] otherppl [dot] com The podcast is a proud affiliate partner of Bookshop, working to support local, independent bookstores. Learn more about your ad choices. Visit megaphone.fm/adchoices
John Keene (winner of a 2018 Windham Campbell Prize for Fiction) talks with Prize Director Michael Kelleher about Mohamed Mbougar Sarr's 2021 Prix Goncourt-winning novel The Most Secret Memory of Men, the joys of a shaggy dog story, the power of the sublime, and the limits of knowledge. Reading list: The Most Secret Memory of Men by Mohamed Mbougar Sarr, tr. by Laura Vergnaud • Blackouts by Justin Torres • Bound to Violence by Yambo Ouologuem • Roberto Bolaño • Clarice Lispector John Keene is a writer, translator, professor, and artist who was named a MacArthur Fellow in 2018. His latest book, Punks: New and Selected Poems, won the 2022 National Book Award for Poetry. In 1989, Keene joined the Dark Room Writers Collective, and is a Graduate Fellow of the Cave Canem Writers Workshops. He is the author of Annotations, and Counternarratives, both published by New Directions, as well as several other works, including the poetry collection Seismosis, with artist Christopher Stackhouse, and a translation of Brazilian author Hilda Hilst's novel Letters from a Seducer. Keene is the recipient of many awards and fellowships—including the Windham-Campbell Prize, the Whiting Foundation Prize, the Republic of Consciousness Prize, and the American Book Award. He teaches at Rutgers University-Newark.
The new Split Braille feature in JAWS 2024 gives Braille display users powerful new functionality, including the ability to view content from different locations on the same Braille line using one of several types of views. When a Split Braille view is active, your Braille display is split into two halves, or regions. Additionally, two vertical lines are displayed between the two regions to show the separation. The use of the Split Braille feature is not restricted to displays of a certain size. It will work with Braille displays of any length though using it with smaller displays may be less beneficial. We have more Split Braille training headed your way, so stay tuned! https://support.freedomscientific.com/downloads/jaws/JAWSWhatsNew
In today's episode of Tech Talks Daily, we embark on an enlightening journey into the complex yet crucial world of designer-developer relationships. Our special guests, Lauren Andres, a Designer Advocate, and Jenny Lea, a Software Engineer from Figma, join us to unravel the intricacies of this dynamic through the lens of Figma's groundbreaking innovation, DevMode. The conversation opens with exploring the traditional challenges that have long defined the designer-developer dynamic. With their rich backgrounds in design and development, Lauren and Jenny shed light on the often fraught nature of this relationship. They emphasize how differing perspectives and languages can lead to misunderstandings and inefficiencies in the collaborative process. This is where DevMode comes into play. Lauren and Jenny delve into how this new tool from Figma revolutionizes how designers and developers interact and collaborate. DevMode, they explain, is not just a technological solution but a transformative approach to aligning language and objectives. It facilitates a shared understanding and a cohesive workflow, enabling designers and developers to work harmoniously while keeping their audience's needs at the forefront. The conversation takes a deeper dive into our guests' personal experiences and learning journeys. Jenny shares her insights on the importance of continuous learning and adaptability in the fast-paced tech industry. She talks about her journey of self-education and growth, providing valuable takeaways for our listeners navigating similar paths. We also touch upon the practical implications of using DevMode in the real world. Lauren and Jenny provide examples from companies like Decathlon and Lanier, showing how DevMode has positively impacted their design and development processes. These stories highlight the tangible benefits of improved collaboration, efficiency, and productivity that DevMode brings to teams. As we explore the human aspect of technology, our guests remind us of the importance of not just relying on tools but also fostering genuine conversations and understanding from the user's perspective. This approach, they argue, is crucial in creating impactful and accessible designs that resonate with users. As the episode draws close, Lauren and Jenny invite listeners to connect with them and learn more about DevMode and its features, including the much-anticipated Annotations function. They also emphasize the Figma team's commitment to user feedback and the continuous improvement of DevMode. This episode is not just a discussion about a tool but a deep dive into the evolving design and development landscape. It's a testament to how technology like DevMode can bridge gaps, enhance collaboration, and lead to more user-centric and practical solutions in the tech world. Please tune in to this insightful episode for a comprehensive understanding of the transformative impact of Figma's DevMode on the designer-developer relationship and discover how it sets a new standard in collaborative technology.
Episode 151:This week we're continuing with:The Worldview and Philosophical Methodology of Marxism-LeninismWritten for the Vietnamese curriculum and translated by Luna NguyenYou can purchase a copy and support translation of the further curriculum here:https://www.banyanhouse.org/product/ebook-the-worldview-and-philosophical-methodology-of-marxism-leninism[Part 1 - 5]Introduction to the Basic Principles of Marxism[Part 6 - 10]Part I: The Worldview and Philosophical Methodology of Marxism-LeninismChapter 1: Dialectical Materialism[Part 11 - 15]Chapter 2: Materialist Dialectics I. Dialectics and Materialist Dialectics II. Basic Principles of Marxist Dialectics III. Basic Pairs of Categories of Materialist Dialectics 1. Private and Common 2. Reason and Result 3. Obviousness and Randomness[Part 16 - This Week]Chapter 2: Materialist Dialectics III. Basic Pairs of Categories of Materialist Dialectics 4. Content and Form - 00:24 a. Categories of Content and Form - 00:24Annotation 150: 0:57 - 17:50 b. Dialectical Relationship Between Content and Form - 17:52Annotation 151: 18:26 - 19:21Annotation 152: 19:39 - 22:25Annotation 153: 22:47 - 25:09 c. Meaning of the Methodology - 25:10Annotation 154: 26:02 - 29:04[Part 17 - 25?]Chapter 2: Materialist Dialectics[Part 26 - 30?]Chapter 3: Cognitive Theory of Dialectical MaterialismFigures:Figure 1 - 3:51A material object can be described in terms of content, inner form, and outer form.Figure 2 - 11:36Figure 3 - 20:30Quantity changes in Content lead to quality shifts in Form.Footnotes:4) 3:26See Annotation 10 and Annotation 108.
No Priors: Artificial Intelligence | Machine Learning | Technology | Startups
Open Source fuels the engine of innovation, according to Arthur Mensch, CEO and co-founder of Mistral AI. Mistral is a French AI company which recently made a splash with releasing Mistral 7B, the most powerful language model for its size to date, and outperforming much larger models. Sarah Guo and Elad Gil sit down with Arthur to discuss why open source could win the AI wars, their $100M+ seed financing, the true nature of scaling laws, why he started his company in France, and what Mistral is building next. Arthur Mensch is Chief Executive Officer and co-founder of Mistral AI. A graduate of École Polytechnique, Télécom Paris and holder of the Master Mathématiques Vision Apprentissage at Paris Saclay, he completed his thesis in machine learning for functional brain imaging at Inria (Parietal team). He spent two years as a post-doctoral fellow in the Applied Mathematics department at ENS Ulm, where he carried out work in mathematics for optimization and machine learning. In 2020, he joined DeepMind as a researcher, working on large language models, before leaving in 2023 to co-found Mistral AI with Guillaume Lample and Timothee Lacroix. Show Links: Arthur's Linkedin Mistral Mistral 7b Retro: Improving language models by retrieving from trillions of tokens Chinchilla: Training Compute-Optimal Large Language Models Sign up for new podcasts every week. Email feedback to show@no-priors.com Follow us on Twitter: @NoPriorsPod | @Saranormous | @EladGil | @ArthurMensch Show Notes: (0:00) - Why he co-founded Mistral (4:22) - Chinchilla and Proportionality (6:16) - Mistral 7b (9:17) - Data and Annotations (10:33) - Open Source Ecosystem (17:36) - Proposed Compute and Scale Limits (19:58) - Threat of Bioweapons (23:08) - Guardrails and Safety (29:46) - Mistral Platform (31:31) - French and European AI Startups
It's YOUR time to #EdUp In this episode, YOUR guest is Dan Whaley, Founder & CEO of Hypothesis YOUR cohost is Dr. Janet Spriggs, President of Forsyth Technical Community College YOUR host is Elvin Freytes YOUR sponsors are The Middle States Commission on Higher Education (MSCHE) & InsightsEDU How can social annotation software increase student engagement & comprehension in the classroom? What are some creative ways that instructors are using Hypothesis to transform their teaching? What does Dan see as the future of Higher Education? Listen in to #EdUp! Thank YOU so much for tuning in. Join us on the next episode for YOUR time to EdUp! Connect with YOUR #EdUp Team - Elvin Freytes & Dr. Joe Sallustio ● Join YOUR #EdUp community at The EdUp Experience! We make education YOUR business! --- Send in a voice message: https://podcasters.spotify.com/pod/show/edup/message
If you've ever been intimidated by Ignatian spirituality, today's guest will put you at ease. In fact, Fr. Michael Hansen is determined to make the Ignatian exercises accessible to as many people as possible. Why? It's simple: He knows that God is at work, whether you're doing the full thirty day retreat or just a one-off reflection. Host Eric Clayton has been intrigued by Mick's work for a while. In fact, longtime listeners will remember when he came on our podcast a few years ago. He talked about the First Spiritual Exercises. Mick wrote a book by the same name. But it's also a shorthand for what Ignatius meant in the 18th Annotation of the full Spiritual Exercises. Both Mick and Ignatius want people to be able to access the riches of Ignatian spirituality, even if you don't have thirty days to sit at a retreat center. Today, Mick brings us up to date on that work. But he also introduces us to a monthlong initiative he's been steering that invites us to contemplate the many Marian devotions of Ignatus of Loyola. Mick and the Jesuits in Australia have produced a beautiful collection of short meditations that pair a Marian image—perhaps a statue that Ignatius would have seen in his childhood or a painting that hung in his room—with guided prayers to help us more intentionally journey through this month of Mary. If you've ever wanted to know what Ignatian contemplation of the Blessed Mothers looks like, this is the episode for you. Learn more about the First Spiritual Exercises: https://jesuit.org.au/ignatian-spirituality/first-spiritual-exercises/ Check out the Marian reflections and get on Mick's mailing list: https://us10.campaign-archive.com/home/?u=75fd6f8cf8deaab17d0961da5&id=3f175b637a
In this episode of Room to Grow, Grace Kelemanik and Amy Lucenta join Curtis and Joanie to talk about how routines can provide the “opportunity and support for each and every student develop mathematical thinking and reasoning.” Although routines are used by most educators for a variety of reasons, Grace and Amy focus on “Routines for Reasoning,” which are specifically designed and structured to surface the ways that students are thinking about the mathematics and to better understanding the reasoning of their classmates to reinforce the mathematics content and thinking goals. In this extended episode, Amy and Grace dive deeply into the “Four Rs” and “Annotation,” two of the five “Essential Strategies” that teachers employ within the routines, with an emphasis on how these strategies provide access and opportunity for all students to engage in the deep thinking of the lesson. Then, they describe the “Connecting Representations” routine in detail to help listeners understand the power of the routines in action. As Grace shares, the power of the routines and essential strategies is that they help teachers to “hand over agency to the students. Teachers are no longer are the sole authority in the classroom... it's the students doing the heavy lifting.” We encourage you to explore the resources below, referenced in this episode:Be sure to explore Grace Kelemanik and Amy Lucenta's website, Fostering Mathematical Practices …...and their books, Routines for Reasoning and Teaching for Thinking. Explore infographics, tasks, and more for the Connecting Representations routine.See the Connecting Representations routine in action in this classroom video.Be sure to join us for part 2 of this conversation next month!Did you enjoy this episode of Room to Grow? Please leave a review and share the episode with others. Share your feedback, comments, and suggestions for future episode topics by emailing roomtogrowmath@gmail.com. Be sure to connect with your hosts on Twitter and Instagram: @JoanieFun and @cbmathguy.
Dr. Joel Biermann of Concordia Seminary-St. Louis Luther's Large Catechism with Annotations and Contemporary Applications Wholly Citizens: God's Two Realms and Christian Engagement With the World A Case for Character: Towards a Lutheran Virtue Ethics The Lutheran Witness magazine The post 2023. Moralistic Therapeutic Deism – Dr. Joel Biermann, 7/21/23 first appeared on Issues, Etc..
In this episode, we discuss the resolution at the forthcoming LCMS Convention that is being put forward to commend the new volume Luther's Large Catechism with Annotations and Contemporary Applications for use by pastors, teachers, commissioned ministers, and laity. Today's guest was the main author of an Overture asking for the distribution of this volume to cease and desist. We'll look at his rationale and discuss what can yet be done about it. Convention Workbook First Issue of Today's Business ----more---- Host: Fr. Jason Braaten Special Guest: Fr. Seth Mierow ----more---- Become a Patron! You can subscribe to the Journal here: https://www.gottesdienst.org/subscribe/ You can read the Gottesblog here: https://www.gottesdienst.org/gottesblog/ You can support Gottesdienst here: https://www.gottesdienst.org/make-a-donation/ As always, we, at The Gottesdienst Crowd, would be honored if you would Subscribe, Rate, and Review. Thanks for listening and thanks for your support.
God Given Talent05-31-1971Neville Goddard Tonight is the Law. We are told in the book of Acts that: “God is not far from each one of us, for in Him we live, and move, and have our being.” (Acts 17:28) I would like to change that a little, and say to you that: God is never so far off as even to be near, for nearness implies separation. And God and Man are one. “Man is all imagination, and God is man, and exists in us and we in Him” [Wm. Blake, from “Annotations to Berkeley”] “The Eternal Body of Man is the imagination, and that is God Himself.” [Wm. Blake, from “The Laocoon”] So, He cannot even be near, for nearness implies separation. On this level, you and I can go amuck, go berserk, exercising this same power that created the universe and sustains it. Your own wonderful human imagination is God. That's God! “By Him all things were made, and without Him was not anything made that was made,” (John 1:3) . . good, bad or indifferent.
Dr. Joel Biermann of Concordia Seminary-St. Louis Luther's Large Catechism with Annotations and Contemporary Applications Wholly Citizens: God's Two Realms and Christian Engagement With the World A Case for Character: Towards a Lutheran Virtue Ethics The Lutheran Witness magazine
William of Ockham is best known today for the model of problem solving known as Ockham's (or Occam's) Razor. But the event that defined his life was an argument with Pope John XXII. Research: Lieberich, Heinz. "Louis IV". Encyclopedia Britannica, 7 Oct. 2022, https://www.britannica.com/biography/Louis-IV-Holy-Roman-emperor Kilcullen, John. “Ockham's Political Writings.” “The Cambridge Companion to Ockham. Cambridge University Press. 1999. Republished online: http://publications.thebritishacademy.ac.uk/pubs/dialogus/polth.html Britannica, The Editors of Encyclopaedia. "Peter Lombard". Encyclopedia Britannica, 20 Aug. 2020, https://www.britannica.com/biography/Peter-Lombard Gál, Gedeon, O.F.M. "William of Ockham Died "impenitent" in April 1347." Franciscan Studies, vol. 42, 1982, p. 90-95. Project MUSE, doi:10.1353/frc.1982.0011 Lambert, M. D. “THE FRANCISCAN CRISIS UNDER JOHN XXII.” Franciscan Studies, vol. 32, 1972, pp. 123–43. JSTOR, http://www.jstor.org/stable/44000287 Donovan, Stephen M. “Bonagratia of Bergamo.” Catholic Encyclopedia. https://www.catholic.com/encyclopedia/bonagratia-of-bergamo Nold, Patrick. “Pope John XXII's Annotations on the Franciscan Rule: Content and Contexts.” Franciscan Studies, vol. 65, 2007, pp. 295–324. JSTOR, http://www.jstor.org/stable/41975430 Knysh, George. “BIOGRAPHICAL RECTIFICATIONS CONCERNING OCKHAM'S AVIGNON PERIOD.” Franciscan Studies, vol. 46, 1986, pp. 61–91. JSTOR, http://www.jstor.org/stable/41975065 Spade, Paul Vincent. “William of Ockham.” Stanford Encyclopedia of Philosophy. March 5, 2019. https://plato.stanford.edu/entries/ockham/ Vignaux, Paul D.. "William of Ockham". Encyclopedia Britannica, 24 Aug. 2022, https://www.britannica.com/biography/William-of-Ockham See omnystudio.com/listener for privacy information.
Dr. Charles Gieschen of Concordia Theological Seminary-Ft. Wayne, IN Luther's Large Catechism with Annotations and Contemporary Applications Concordia Theological Seminary Angelomorphic Christology: Antecedents and Early Evidence (Library of Early Christology)
Dr. Joel Biermann of Concordia Seminary-St. Louis Luther's Large Catechism with Annotations and Contemporary Applications Wholly Citizens: God's Two Realms and Christian Engagement With the World A Case for Character: Towards a Lutheran Virtue Ethics The Lutheran Witness magazine
Dr. Jordan Cooper of Just and Sinner Just and Sinner Dr. Cooper's Website
Dr. Jordan Cooper of Just and Sinner Just and Sinner Dr. Cooper's Website
Have you ever wondered where the earliest sermons in the Bible are found? Moses' final speech to Israel, found in Deuteronomy, is the first time we see what is essentially a modern sermon—a long speech meant to communicate God's truth. Just as Israel is about to enter the promised land, Moses reminds them that, just like their ancestors, they have the choice to live by their own wisdom or to follow Yahweh's life-giving commands. Join Tim and Jon as they dive into the final scroll of the Torah and explore the choice before Israel—and the choice we face today too.View full show notes from this episode →Timestamps Part one (00:00-20:00)Part two (20:00-40:38)Part three (40:38-1:02:14)Referenced ResourcesIntroduction to the Old Testament as Scripture, Brevard S. ChildsInterested in more? Check out Tim's library here.You can experience the literary themes and movements we're tracing on the podcast in the BibleProject app, available for Android and iOS.Show Music “Defender (Instrumental)” by TENTS"Praise through the Valley" by Tae the Producer"Happy Scene" by Sam StewartProduced by Cooper Peltz with Associate Producer Lindsey Ponder. Edited by Dan Gummel, Tyler Bailey, and Frank Garza. Annotations for our annotated podcast in our app by MacKenzie Buxman.Powered and distributed by Simplecast.
© 2020-2025 Ivy Podcast Discovery LLC
