Podcasts about Cache

Le Parisien révèle un nouveau scandale dans le secteur des fourrières à Paris. En 2023, l'affaire du « Roi des fourrières » avait mis en cause Chafic Alywan, le patron de l'entreprise Inter Dépannage qui régnait en maître sur la moitié sud des fourrières parisiennes. Cet homme d'affaires libanais proche de l'ancien ministre Olivier Stirn est toujours en attente de procès pour corruption, faux et blanchiment, ainsi qu'une dizaine de policiers et un ancien fonctionnaire. Principalement soupçonné de revente illégale de véhicules immobilisés, l'ancien patron sera aussi questionné sur des montages financiers et du travail non déclaré.Selon une enquête interne consultée par le Parisien, il pourrait avoir profité du manque de contrôle de la Ville de Paris sur le secteur des fourrières. Un problème endémique qui ressurgit aujourd'hui alors que certains demandent des mesures pour son assainissement. Comme le révèle Le Parisien, la société qui a pris la suite d'Inter Dépannage, DLA, fait elle aussi, de nouveau l'objet d'une vaste enquête de gendarmerie. Récit d'un double scandale dans cet épisode de Code Source avec Nicolas Jacquard, journaliste au service Police-Justice du Parisien qui révèle cette nouvelle affaire dans nos colonnes.Écoutez Code source sur toutes les plates-formes audio : Apple Podcast (iPhone, iPad), Amazon Music, Podcast Addict ou Castbox, Deezer, Spotify.Crédits. Direction de la rédaction : Pierre Chausse - Rédacteur en chef : Jules Lavie - Production : Clémentine Spiler - Réalisation et mixage : Julien Montcouquiol - Photo : Jean-Baptiste Quentin - Musiques : François Clos, Audio Network. Hébergé par Acast. Visitez acast.com/privacy pour plus d'informations.

durée : 00:52:14 - Grand bien vous fasse ! - par : Eva Roque - Un engouement pour l'ultra-trail et les courses toujours plus extrêmes, l'essor de l'IA qui nous promet d'être plus rapide et efficace, des agendas remplis, et des nuits de sommeil qui s'écourtent. Pourquoi cherchons-nous toujours plus d'intensité dans nos vies ? - équipe : Matthias Volant, Anna Massardier, Jessica Bagic, Manon Latour, Johanna Houssin - invités : Victoire Tuaillon Journaliste et autrice française, Olivier Bessy Professeur de sociologie émérite à l'Université de Pau, spécialiste du sport et des loisirs Vous aimez ce podcast ? Pour écouter tous les épisodes sans limite, rendez-vous sur Radio France

What happens when a gadget cache doesn't just have one unusual attribute, but a combination so rare you may have never seen it before? In this episode of Gadget Talk, we dig into the world of rare gadget cache attribute combinations and explore how creative cache owners can use attributes to make a hide more distinctive, informative, and memorable. From unexpected pairings to unusual technical requirements and “how do those two even go together?” combinations, we'll look at what makes certain attribute stacks stand out. We'll also talk about the difference between choosing attributes because they're rare and choosing them because they genuinely describe the experience. Can an unusual combination inspire an entirely new gadget-cache concept? Which attributes create interesting design challenges? And how can builders make sure the cache delivers on the expectations those little icons create? Subscribe to Geocache Talk on YouTube: https://www.youtube.com/GeocacheTalk Check out more of the Geocache Talk Network of Shows here: https://geocachetalk.com/ https://www.facebook.com/geocachetalk https://twitter.com/geocachetalk https://instagram.com/geocachetalk geocachetalk@gmail.com #geocaching #geocachetalk www.youtube.com

Écharpe autour du cou, la jeune fille noircit inlassablement les pages d'un carnet d'un air songeur. Jade, c'est son prénom, évoque les héroïnes des studios Ghibli. Parfois, elle jette son regard distrait vers sa fenêtre, où un gros chat roux squatte le rebord. Cette séquence d'animation de 20 secondes se répète à l'infini depuis 2017, l'année de naissance de la chaîne Youtube Lofi Girl. La vidéo n'a ni début, ni fin, pourtant, environ 40 000 personnes la visionnent en simultané. Le détail le plus important de cette courte vidéo n'est pas ce qu'il se passe à l'image, mais la musique qui passe. Car, détail important, Jade, porte sur ses oreilles un gros casque diffusant des beats lofi.  Mais alors, comment cette vidéo a-t-elle été créée ? Écoutez la suite de cet épisode de "Maintenant Vous Savez - Culture". Un podcast Bababam Originals, écrit et réalisé par Jonathan Aupart. Première diffusion : décembre 2023 À écouter aussi : ⁠Quel est le lien étonnant entre Walt Disney et la Normandie ?⁠ ⁠Que sont devenus les gagnants de la Star Academy ?⁠ ⁠Quels sont ces films avec des scènes de sexe non simulées ?⁠ Retrouvez tous les épisodes de "Maintenant vous savez". Pour rester informé et recevoir le meilleur de “Maintenant vous savez” chaque semaine, abonnez-vous à notre newsletter. Suivez Bababam sur Instagram. Learn more about your ad choices. Visit megaphone.fm/adchoices

On me dit à l'oreillette qu'en plus d'amortir ton abonnement au Club HOURRAIL ! grâce au bon d'achat GreenGo de 34 €, les membres auront droit, dès le 26 août, à de nouveaux avantages exclusifs.

On this episode of the Cedric Maxwell Podcast, hosted by CLNS Media's Nick Gelso and Gary Tanguay, Celtics legend Cedric Maxwell doesn't hold back on the Jaylen Brown trade — the criticism Jaylen's taken since leaving, what a championship in Philadelphia would mean for his legacy, and why Boston's new cap situation has him worried about the Celtics' next move. Plus: comparing the Jaylen Brown fallout to the Luka Dončić trade reaction, and why Jaylen picked up Marcus Smart's mantle as the Celtics' enforcer. 0:00 - Cold Open: “I'm Pissed Off” — Jaylen Brown trade reaction 0:35 - Sixers chemistry: who's playing for a contract? 2:11 - Predicting Jaylen's role in Philly by January 3:07 - Cache debate: LeBron vs. Embiid vs. Jaylen 4:38 - PrizePicks 6:49 - Cedric Maxwell: “I hated to see him go” — Jaylen's legacy 7:53 - Cap aprons are dismantling super teams 8:44 - Dynasties vs. parity: what Celtics fans actually want 10:24 - Jaylen Brown trade vs. the Luka Dončić fallout 12:10 - The pressure now on Jayson Tatum 12:54 - The “smartest guy in the room” quote controversy 13:50 - Back to “Pissed Off” — the comments made about Jaylen 14:28 - Marcus Smart & Jaylen Brown's enforcer legacy The Cedric Maxwell Podcast on CLNS Media is Powered by:

Cache Council approves members of Rec Center Feasibility Study Committee -- Logan Council approves property tax increase -- 

À l'occasion de sa revue de presse, vendredi, Paul Arcand commente un dossier de La Presse qui révèle qu'entre 2020 et 2024, le nombre de femmes en situation d'itinérance ayant perdu la vie a bondi de plus de 200% dans la province. Ces chiffres, qui proviennent d'une étude de l'Université de Sherbrooke, indiquent également que plus du tiers des personnes qui sont décédées durant ces quatre années se trouvaient en situation d’«itinérance cachée» au Québec. Ces données alarmantes, basées sur le dépouillement des dossiers du Bureau du coroner, ne brossent toutefois qu'un portrait partiel de la réalité. Les chercheurs rappellent que leurs travaux se concentrent uniquement sur les morts violentes ou suspectes, laissant présager un bilan réel encore plus lourd. Autres sujets abordés Les promesses électorales; Vincent Guzzo est accusé de voies de fait; Ce sera le bordel en circulation à Montréal pendant tout le mois de septembre. Voir https://www.cogecomedia.com/vie-privee pour notre politique de vie privée

On s'immerge dans la genèse de la célèbre chanson Dors Caroline. L'auteur Pierre Flynn, lève le voile sur les origines inattendues de cette œuvre poignante. Inspirée d'une enquête troublante menée dans les rues de Brooklyn, la pièce a abordé des thèmes sombres de manière totalement imprévue. Comment ce récit sur la jeunesse en détresse est-il devenu un succès phénoménal contre toute attente ? Voir https://www.cogecomedia.com/vie-privee pour notre politique de vie privée

durée : 00:44:14 - Questions du soir : le débat - par : Antoine Dhulster - Que nous cache encore la Lune ? On en parle ce mercredi 12 août 2026 à quelques minutes d'une éclipse solaire historique. De la physique lunaire aux enjeux de l'exploration de notre satellite, une petite leçon d'astronomie pour tous avec nos invités. - équipe : Phane Montet, Louise Cognard, Tom Umbdenstock, Inès Bouffartigue Sebastia, Adèle Triol, Coline Guillot - invités : Guillaume Hébrard Directeur de recherche CNRS à l'Institut d'Astrophysique de Paris et à l'Observatoire de Haute Provence, Sylvain Bouley Planétologue et professeur au laboratoire GéoSciences Paris Saclay (GEOPS). Vice-président de la Société Astronomique de France., Nicolas Beck Vulgarisateur, délégué régional académique adjoint à la recherche et à l'innovation Grand Est Vous aimez ce podcast ? Pour écouter tous les épisodes sans limite, rendez-vous sur Radio France

Un leader solide ne se reconnaît pas à son titre, mais à son cœur. Découvre les 7 signes qui révèlent une insécurité cachée dans le leadership et apprends à devenir un leader libre, humble et profondément enraciné dans les principes du Royaume de Dieu.➡️➡️ FORMATION GRATUITE             LIBÈRE LE LEADER EN TOI            CLIQUEZ ICI ➡️ https://exponentiel.net/formationleaderLIENS ET RESSOURCESFormation GRATUITE pour développer ton potentielhttps://exponentiel.net/potentielRejoins notre programme de MENTORAThttps://exponentielclassroom.com Service de COACHING personnaliséhttps://ecoachingexponentiel.comNos nouvelles formationshttps://formations.exponentiel.net/formationsNotre chaine youtube - ExponentielTVhttps://www.youtube.com/@ExponentielnetNotre site web: https://exponentiel.net/Instagram:https://www.instagram.com/exponentiel... https://www.instagram.com/lucdumontof... 

Dans ce nouvel épisode du "Journal Imprévisible", Marc Bourreau nous plonge dans l'univers fascinant d'Agatha Christie, l'une des plus grandes autrices de romans policiers du XXe siècle. À travers des archives et des témoignages, il retrace la vie extraordinaire de cette femme hors du commun, dont les œuvres regorgent de rebondissements et de mystères.Dès le début, on est saisi par les détails incroyables de la vie d'Agatha Christie. Son enfance atypique, marquée par une mère passionnée de spiritisme, son apprentissage autodidacte de l'écriture, son talent pour maîtriser les subtilités du poison - tout semble avoir contribué à forger l'imagination débordante de cette reine du crime. Mais au-delà de ses romans à succès, c'est la personnalité même d'Agatha Christie qui intrigue. Son mystérieux séjour de 11 jours dans un hôtel sous une fausse identité, ses voyages qui ont inspiré certains de ses plus célèbres décors de crime, tout cela donne l'impression que la vie de la romancière était aussi captivante que ses propres intrigues. Marc Bourreau nous fait également découvrir les secrets de l'écriture d'Agatha Christie, de sa méthode de travail rapide et efficace à sa formule gagnante, mêlant toujours habilement la question "Qui a tué ?" à celle du "Pourquoi ?". On ne peut qu'être impressionné par cette productivité hors norme, qui a fait d'elle l'autrice la plus lue au monde après la Bible et Shakespeare.Hébergé par Audiomeans. Visitez audiomeans.fr/politique-de-confidentialite pour plus d'informations.

Ta balance ment. Le vrai chiffre qui compte, c'est pas celui-là.Avec Keven Bouchard qui travaille au développement des affaires avec InBody, on parle de gras viscéral, de masse musculaire pis de pourquoi tu peux "avoir l'air en forme" et quand même avoir un problème de santé qui s'en vient.Pour l'occasion, on fait tirer une balance résidentielle InBody. Le tirage se fait le 14 août 2026.Lien vers le concours

A partir de mars 2022, Kana réédite le fameux Naruto en grand format pour les 20 ans du manga en France. Une série historique pour l'éditeur français. L'occasion de retracer le parcours de son auteur Masashi Kishimoto, et de voir qu'avant le succès faramineux qu'on lui connaît, la route a été longue. Oui, on peut parler d'un parcours, digne d'un héros de shônen. Mais d'où lui vient la passion du dessin ? Quelles études a-t-il réalisées ? Et comment le manga Naruto a-t-il explosé ? ⁠Ecoutez la suite dans cet épisode de "Maintenant vous savez - Culture".⁠ Un podcast Bababam Originals, écrit et réalisé par Jonathan Aupart. Première diffusion : février 2022 A écouter aussi : ⁠Quelle est l'incroyable histoire du hip-hop ?⁠ ⁠Qu'est-ce que c'est le rococo ?⁠ ⁠Quels sont ces films maudits du cinéma ?⁠ Retrouvez tous les épisodes de "Maintenant vous savez". Pour rester informé et recevoir le meilleur de “Maintenant vous savez” chaque semaine, abonnez-vous à notre newsletter. Suivez Bababam sur Instagram. Learn more about your ad choices. Visit megaphone.fm/adchoices

For Select 399, Cairo-based DJ Hashad delivers a journey through the different shades of electronic music, moving from deep, groove-led rhythms into more energetic territory while maintaining a steady sense of movement throughout. Known for his fluid approach to house music, Hashad's sets draw from a wide range of influences, shaped by years spent navigating Egypt's nightlife scene and understanding the dynamics of the dancefloor. His selections are built around momentum and storytelling, carefully balancing atmosphere, rhythm and energy from start to finish. Having shared the decks with acclaimed names including Traumer, D Stone and Apollonia, Hashad continues to develop a sound that reflects both his local roots and his evolving relationship with electronic music. Starting with a deeper groove and gradually pushing into more energetic territory, this set features tracks including Nyra's ‘Energy Bliss', DC Salas' ‘A Departure', Dart and Kara Okay's ‘Can You Hear Me' and Cache and Mella Dee's ‘Don't Hurt Me (Mella Dee Full Pump Mix)'.

Keegan Garrity on Cache Council rec center committee vote -- Cache Water District board member Jeannie Simmons

In this episode of PING, we talk with Job Snijders about secure Internet routing again, focusing on an approach to preserving the state of RPKI as a longterm historical record for research and analysis. Job was last on PING to discuss the “Erik” protocol. Job has been running an archive of RPKI state for some time, as a volunteer activity but an exercise which was feasible when BGP speakers producing ROA objects was measured in the hundreds to low thousands is significantly more expensive when the population of ROA producers is a more realistic percentage of the around 80,000 AS holders worldwide. Because Public Key Cryptography depends on a regular re-signing, and re-cataloging (as another signed object, the RPKI “Manifest” file) even when there is no substantive change in the state of signed information, a large amount of “churn” can be seen in the data, and even a compressed form of this state in turn incurs a huge overhead in storage of hard-to-compress data. Job reached the limits of his free activity, and started to explore a more compact and useful representation. This has emerged in the IETF standards process as two related activities. The Canonical Cache Representation or CCR, and the aggregation over this for data represented as the RPKI Spool data model for “materialising” the state of RPKI objects. Along with his fellow IETF draft authors Bart Bakker , Tim Bruijnzeels, Theo Buehler and Fedor Vompe Job has managed to define a remarkably compact, highly compressible representation of RPKI validated objects, and the cryptographic payloads. These systems have been designed to leverage well known UNIX and other techniques for data management such as the “tar” format for spooled data, ASN.1 for binary encoding, Merkle Trees to hold compact hash state, and data compression using the zstd encoding. It's a nice approach to solving the scaling problem.

Watch the full episode on YouTube:We first covered Baseten last year when DeepSeek mania was at peak hype. Now they have raised a monster $13B round and become one of the new cohort of AI Infra decacorns that are (with Nvidia, Intel, and the semis complex) chief beneficiaries of the Inference Inflection. We return to Baseten at the peak of the 2026 edition of Open Weights debate. Ali has published a viral breakdown of Kimi K3:And since you last saw him, Philip has spoken at AI Engineer and written the definitive book on Inference Engineering spotted all over SF:Three years ago, inference engineering barely existed as a category.Today, it is one of the most critical disciplines in AI. Inference engineering inherently tackles a different question than standard model training: “How do you turn those weights from training into a product that is fast, reliable, and affordable at scale?” Focusing on these creates an entirely new optimization problem.In one recent GLM-5.2 experiment, quantizing more of the model actually preserved its benchmark quality while increasing throughput by 20%, because the errors introduced in different layers could cancel each other out.Inference is no longer just the final step after training. It is becoming its own engineering discipline, with its own research problems, infrastructure, and increasingly specialized roles.In this episode, Baseten's Philip Kiely and Ali Taha join swyx and Vibhu to explain what actually happens after a new open model is released and what it takes to turn “we generated a token” into a fast, reliable, production-ready API.We go deep on cache-aware routing, disaggregated prefill and decode, quantization, speculative decoding, KV-cache movement, model parallelism, GPU kernels, and the race to make frontier models up to 10× faster. Philip and Ali explain why inference optimizations can still produce gains of 20%, 100%, or even 200%; how quantization errors can cancel one another out; why identical weights can behave differently across clusters; and how Baseten grafted a Kimi vision encoder onto GLM-5.2 without changing the underlying language model.The conversation then expands beyond LLMs into NVIDIA Dynamo, mega kernels, Rubin, AI-specific chips, local inference, video generation, diffusion versus autoregressive models, and the enormous compute barrier to generating coherent long-form video. Finally, we explore the convergence of training and inference, continual learning through persistent KV cache, and the emerging loop where models help optimize the infrastructure that runs them.We discuss:* What happens when a 200,000-token request enters an inference system* Cache-aware routing and reusing previously computed KV cache* Why prefill and decode are increasingly handled by different GPUs* When dedicated deployments become cheaper and more reliable than shared APIs* How speculative decoding uses a smaller model to accelerate a larger one* Tool calling, structured outputs, and what LLMs actually do* What it takes to support a new open model on day zero* Grafting Kimi's vision encoder onto GLM-5.2* Retrofitting inefficient model layers with components from other architectures* Why models sometimes collapse into repeating the same token* How hardware, kernels, and race conditions create nondeterministic failures* Preserving model fidelity while making inference faster* How quantization errors can cancel each other out* Why inference optimizations still deliver gains of 20%, 100%, and 200%* How optimized serving can make a model up to 10× faster* NVIDIA Dynamo, KV-aware routing, and distributed model serving* Speculative decoding the speculative decoder* Why local AI is about making models less dumb while data-center AI is about making them less slow* Tensor, expert, and pipeline parallelism across GPUs* Hardware-aware model design, auto-tuning, and the case against mega kernels* Rubin and why inference is becoming a systems problem* Whether modern GPUs are evolving into programmable AI ASICs* Why enormous models like Kimi K3 require GB300-class hardware* Why open-source video generation still trails Veo, Kling, and other closed models* The quadratic attention bottleneck behind long-form AI video* Autoregressive video, real-time generation, and compounding quality drift* Why future video systems may combine autoregressive and diffusion architectures* Training for inference and inference for training* Continuous post-training, deployment, evaluation, and improvement loops* How GLM-5.2 helped optimize the kernels serving GLM-5.2 itself* Why faster networking could unlock dramatically faster decoding* Continual learning, KV-cache compaction, and persistent model memoryShow Notes* How to build a day-0 API for Kimi K3* 22580: From GPT2 to Kimi3, ExplainedPhilip Kiely* LinkedIn: https://www.linkedin.com/in/philipkiely* X: https://x.com/philipkiely* Inference Engineering: https://www.baseten.co/inference-engineering/Ali Taha* LinkedIn: https://www.linkedin.com/in/aliestaha/* X: https://x.com/waterloointernTimestamps00:00:00 Introduction and the 200K-Token Prompt00:03:18 Dedicated Deployments, Speculative Decoding, and Tool Calling00:11:26 Launching Production-Ready Open Models00:19:06 Model Retrofits, Failure Modes, and Nondeterminism00:28:22 Quantization and Canceling Errors00:32:15 The Race to 10× Faster Inference00:40:48 Dynamo, Speculation, and Local vs. Data-Center AI00:50:18 Model Parallelism, Auto-Tuning, and Mega Kernels01:00:55 Rubin, GPUs vs. ASICs, and Custom AI Chips01:10:03 Giant Models and the Limits of GPU Memory01:12:42 AI Video, Quadratic Attention, and Autoregressive Generation01:21:47 Audio, Images, and Diffusion Models01:27:32 Training, Self-Optimizing Models, and Continual Learning01:40:06 Closing ThoughtsTranscriptIntroduction: Baseten, Waterloo Intern, and Inference EngineeringSwyx [00:00:00]: Okay, we're here in the studio with Philip, old friend from Inference Engineering, the book, as well as Baseten and everything that you've done, you and I have done before, as well as Ali. Welcome.Ali [00:00:15]: Pleasure to meet you.Swyx [00:00:15]: Waterloo intern.Ali [00:00:16]: Waterloo intern, always.Swyx [00:00:17]: When did you get “Waterloo intern” as a handle?Ali [00:00:19]: As a handle? Oh.Ali [00:00:20]: I think the rebranding happened mid-March. When I saw it was open, I was like, “I have to take it. Up for grabs.”Philip [00:00:26]: The problem is that Ali is really good at his job and is not gonna be an intern much longer.Philip [00:00:30]: So we have to figure out who's gonna get the handle.Ali [00:00:33]: Well, I'll pass the torch over to the next intern.Swyx [00:00:34]: Oh, okay. It can be, like, you just pass it to another Waterloo grad.Ali [00:00:37]: To another Waterloo intern. No, bruh.Philip [00:00:39]: Yeah.Ali [00:00:39]: Intern.Swyx [00:00:40]: Intern, yeah.Ali [00:00:40]: And no.Philip [00:00:41]: You gotta get an intern from Waterloo.Ali [00:00:42]: Yeah, I've gotta get an intern from Waterloo.Swyx [00:00:44]: Right.Ali [00:00:44]: But they have to follow the path.Swyx [00:00:45]: Oh, it could, but it could come from Baseten, so it's like whoever Baseten gets from Waterloo.Ali [00:00:48]: Right.Swyx [00:00:49]: Has the title of Waterloo.Ali [00:00:50]: It stays in the ecosystem.Philip [00:00:51]: Exactly.Ali [00:00:52]: Halfway through the internship, you either get it or you're out.Philip [00:00:55]: You should also do, like, a big graduation ceremony where you change the handle.Ali [00:00:59]: Just say it.Philip [00:00:59]: For everybody.Swyx [00:01:00]: You guys are good at ceremonies, clearly. We had a nice launch of the book, very successful. But before we get into all that, I wanna start off with a fun question for you. Okay, you're an expert inference engineer. What happens when I send a long query, say two hundred thousand tokens into Baseten's inference? What's the process of query through GPU model routing, balancing, all that? What is all the stuff that we don't think about?Long Context Requests, KV Cache, and Cache-Aware RoutingPhilip [00:01:26]: With a long query specifically, the first thing that I'm gonna ask is, “Have you sent me this query before, or at least part of it?” and I really hope you have, because it's gonna be a lot easier for me and a lot cheaper for you. So the first thing that we're gonna look at is some cache-aware routing, where we're going to see, we probably have a number of instances, a number of replicas up serving whatever model you're hitting. We want to send this one to something with, number one, available prefill workers, and number two, ideally some cached input already there so that we can skip prefill on at least part of these two hundred thousand tokens. If you're doing two hundred thousand tokens, it's probably coding or a multi-turn agent or something where you would expect to have that cached. If you don't, we're gonna have to send it to a prefill worker. We've at least on certain models disaggregated prefill and decode, so you're going to have one set of GPUs that's solely going to process the input, create the KV cache, and get you your first token, and then that's going to be passed over to a separate set of GPUs, which is going to run decode. We're going to iteratively make those tokens. We're probably going to have some speculator model in front of that. I'm going to assume that you're doing coding, and because of that, our speculator model, which assumes you're doing coding, is gonna have a high draft token acceptance rate. If I'm wrong and you're asking me to summarize every Harry Potter book, it's gonna be slower. And then we stream that output to you and account for it, charge you, a couple of pennies and say, “Hey, would you like to send another one?”Swyx [00:03:04]: Except Baseten doesn't charge by pennies.Philip [00:03:07]: Well, yeah, we charge. I'm assuming that we're talking about the public model APIs. If you are setting up a dedicated deployment, then yeah, it's not pennies.Public APIs vs. Dedicated DeploymentsSwyx [00:03:18]: Yeah, one of the key differentiators when I was talking with Baseten initially was that people who want very high volume just need to rent by the box, ‘cause then it's up to you to figure out how to saturate the box.Ali [00:03:31]: And more often than not, it's, like, way cheaper if you're pushing, like, millions of tokens per hour, if you just pay per hour instead of pay per token.Philip [00:03:37]: Yeah, they do. I think that we've increasingly seen a lot of demand for the pay per token APIs, just because everyone wants to try open models, and then once they find a use case that's really sticky, then they move over to dedicated.Swyx [00:03:51]: Is there a best practice on when it's time to swap over?Philip [00:03:54]: Couple reasons. Yeah, reliability, that's a big one, right?Ali [00:03:57]: Like, if they have a very specific use case, they want you to train something specifically for them, like they want their own spec dec, for instance, for their own traffic.Swyx [00:04:04]: Spec dec is speculative decoding.Speculative Decoding and Custom SpeculatorsAli [00:04:05]: Speculative decoding, yeah.Swyx [00:04:07]: You have to explain.Ali [00:04:07]: Sorry. Like, speculative decoding is like, if you have a huge model, right? And so the model is going to be generating one token at a time every single turn, every single forward pass. So we attach, like, this little, like, parasite, like this layer that goes on top of the model, and this model just has to predict. It does three very fast autoregressive forward passes, and it will predict, like, three certain tokens, and then you do one forward stage over the entire original model in order to see if those predictions were correct or not, and then you accept them or you reject them. Now, this draft model is traffic specific, so if you, like, Philip said, if you're summarizing Harry Potter books, I can train exclusively that draft model on Harry Potter books, and I can guarantee you that I'm gonna accept the three tokens every single time. And so with that case, I increase your decode speed. I wouldn't be able to provide this to you if you're a shared endpointSwyx [00:04:53]: YeahAli [00:04:53]: ‘cause I have no idea if you're doing Harry Potter, if you're doing coding, if you're doing English. We don't know. Also, there was a thing in the book that mentioned that if they really cared about a specific threshold, chapter four, I think. Do you remember that?Philip [00:05:06]: Yeah. The things that you can do is you can set a specific, like, batch sizing, a specific, like, parallelism strategy if you're trying to optimize for, like, throughput versus latency. You can. Maybe a NVFP4 quant doesn't pass your benchmarks and you wanna run a model at higher precision, you could do that. There's just a bunch of reasons why you might wanna have your own endpoint and the biggest one, of course, just being, like, you don't have to deal with someone else doing a hundred million tokens of benchmarking traffic at the endpoint when you happen to be trying to serve your users.Swyx [00:05:40]: Yeah. I think one thing that is. That is a classic journey. Like, it's people is asking the, what happens when you type Google into the browser. Tool calling, is that just, you're generating JSON or is there more complication beyond that?Tool Calling, JSON, and Structured OutputsAli [00:05:58]: Certain customers that we have, they have their own post-trained models, and so they demand a tool calling that's not just, like parse a file or go find the weather. It's something that's very specific and you have to do post-training on this. And if the post-training on the model is not good or if the quantization after the post-training to get the inference to be fast, the model will struggle reading the JSON file and reading the tool calling. But it doesn't require its own like sandbox. It's not like it's going to use that tool calling to like escape a sandbox or like it doesn't have to be contained. It can just be a normal dedicated deployment. The challenge with tool calling more and more seems to be that the companies want certain tool calling which is a very sensitive thing to train. And because you're dealing with all of the JSON outputs, if it doesn't like close the end of the request in a very certain manner, you end up with a model that did the tool calling and like the thinking and so as a result of that, it didn't see the result and just hallucinated the result as it decoded. That seems to be the most challenging thing with tool calling, not really the sandboxes model.Philip [00:06:56]: Yeah, that's a challenge on the training side and then on the inference side, there's work that you can do to scope the possible output. So we published this at this point close to two years ago, the solution to this problem which is you make a state machine and you use that to constrain the output to a specific format. So this is the structured output problem. If you remember backSwyx [00:07:27]: Yeah, the specific grammar is,Philip [00:07:29]: Yeah, exactlySwyx [00:07:30]: GML had this thing.Philip [00:07:31]: Yeah. So it's like the old-school “make sure this is only JSON”, return only JSON orSwyx [00:07:38]: YeahPhilip [00:07:38]: Grandma's gonna die type of prompts.Swyx [00:07:39]: Is it BNF grammar? At some point OpenAI had released a thing that was like, yeah, if you want to constrain your output, write BNF grammar, back as NOR.Philip [00:07:47]: In our inference system, it's just a specified output format. And you get the guarantee that your output's gonna be structured along that format. And so applying that to tool calls can like help cut down on. You can still call the wrong tool or call no tool. It doesn't solve the certainty problem but it at least solves the output structuring problemSwyx [00:08:10]: YeahPhilip [00:08:10]: Within tool calls.Swyx [00:08:12]: And MCP is just another form of tool, right.Philip [00:08:14]: Yeah, exactly.Swyx [00:08:15]: As far as there's no special thing there.Philip [00:08:16]: The thing I'm always like explaining to people is the LLM is not capable of doing anything. It's only capable of making suggestions of what to do and then if those suggestions are formatted in a certain way and applied to a system that knows what to do with them, then an action occurs.Swyx [00:08:32]: Yeah. Part of the fun stuff is, this is solved outside of tool calling too. Like in an agent loop if the output is not correct or you're right, like reasoning, tool calling was done in the reasoning trace, just be like, “Oh, I don't know what to do. Let me just try again.” And it might get there after a few tries. And on your point of training, sometimes this is harder in smaller models, so you don't have the same exact quality outputAli [00:08:56]: Right.Swyx [00:08:57]: When you just swap from a big model, right?Ali [00:08:59]: Yeah. I will say that, before, I think we need to go back to inference engineering proper.Ali [00:09:04]: But, I had expected that something would replace JSON because it's hard to stream JSON ‘cause JSON must be complete and you must have open and close brackets and everything. So it's hard to parse something or validate something while it's being streamed. So people invented all sorts of things that are like, I forget the name of some of these alternatives, but it's something like TOML, something like YAML. But JSON seems to be dominant still.Philip [00:09:30]: The JSON outputs aren't that long, right? Like you could have a long-- ‘cause tool calls also contain the arguments in them and perhaps for a certain tool you might pass like a very long argument. But my impression of the median tool call is that it's a relatively small number of tokens, right? So I would expect that speculators are generally fairly good at something as formatted as JSON. And so you would have like a pretty fast decode step there and that the streaming wouldn't be as valuable, but maybe I'm wrong about that.Ali [00:10:02]: I think you're also bounded by the software or that the model is gonna integrate with if the software is built with JSON for the tool calls or if the company that you'- if your customer says that this is how our software works and our tools are interfaced with JSON, you can ask them to like, change their software and say like, “Yeah, this is gonna be better for the model.” but like with the right training shouldn't be that much of a difference. Also more profitable if it outputs more tokens probably.Swyx [00:10:25]: Depends on your business model.Swyx [00:10:27]: It really depends. But I will say that, as a writer with like experience a lot with generated output, I do try to move from text to JSON text which is very long JSON, right? Like there's paragraphs in every field because I'm trying to structure it, right?Philip [00:10:44]: Right.Swyx [00:10:44]: I want you to first make factual statements, then make opinions then make bullet point summaries, have dates, have entity references have your sources for references, all these things. Anyway, so these are things that like I think people who really experiment with structural output have to really care about. But, let's, let's recurse up the stack a little bit. Before we started recording, you mentioned something really cool, which is that there's a lot of engineering that-- inference engineering that goes on when a new model provider releases a new model, right? So let's call it GLM-5.2, Kimi K3. I had previously assumed, especially if it's like, well, GLM 5 to 5.1 to GLM-5.2, like that you've supported them before. Is it that much work?What It Takes to Support a New Open ModelAli [00:11:26]: It's a lot of work.Swyx [00:11:28]: Yeah. Okay. So like, a lot of people, all you guys, right whenever a new model launch like, people rush to say like, “Oh, Hugging Face supports this, Fireworks supports this, Spacetime supports this,” and I'm like, “Yeah, of course we support it.” But what goes into that? What goes intoPhilip [00:11:40]: I think it's more than just support it too, right? It benefits the consumer a lot. Like I think it was with Kimi K2.5 or GLM-5.2 the latest, there was an inference war, right? X provider is at 90 tokens a second. The next day we're at 150. The nextSwyx [00:11:55]: I kinda kicked that off with the GLM-5.2.Swyx [00:11:58]: I wrote a Twitter article about. It got like half a million views,Ali [00:12:02]: Based on being numberSwyx [00:12:03]: YeahAli [00:12:04]: Or it's for something else.Swyx [00:12:05]: Yeah. Which,Ali [00:12:06]: Oh my GodSwyx [00:12:07]: Which then got everyone really excited about, hey, how can we, bend tracks a little bit further and,Philip [00:12:14]: There's a difference between support the model, as in I can make a token out of this model, and support a model, as in I have a production-ready API from this model.Philip [00:12:26]: Getting to the point of I can make a token out of this model is not that hard because generally the, open source inference engines, vLLM, SGLang of the world oftentimes even receive weights ahead of time, maintainers do, or the people making the model merge PRs to ensure support. So you generally can, just get it working on the standard open source stack without too much pain in most cases. The challenge is, every inference company is gonna have own proprietary stack. Some open source components, some in-house stuff. And for any arbitrary model, there's going to be some new stuff. Sometimes you get lucky, like K, two five to two six was, like, pretty similar.Quantization, Speculators, and Production ReadinessAli [00:13:16]: Yeah. It was pure continued post-trainingPhilip [00:13:18]: YeahAli [00:13:18]: If I remember correctly.Philip [00:13:19]: Even in those cases, there's still stuff you have to do. You have to redo the quantization work. You're taking the model from. Generally, these models are not released in NVFP4, and we want them to be in NVFP4 for maximum Blackwell compatibility. So we have to perform that quantization, and, calibrate the quantization to make sure that we're not causing any regression in the model's intelligence. And then we also have to train the speculator, as we've talked about. Generally, we have. We have ZDR, zero data retention on our model APIs, so we don't know exactly the traffic that people are sending us, but we know what's popular. We know that coding use cases are popular. We know that agents, agentic use cases are popular. So we can get public data sets that are representative of that traffic and train general speculators. Now, with speculators today, you need to train the speculator using the base model itself because you're getting hidden states out of the model from running inference on these specific prompts, and that is the training data you use to create the speculator. So there's that process which you need the real model weights for. And then there's of course just the process of, standing up all the infrastructure behind it, loading all this stuff, testing it. And then when there's a new model with a newer architecture, I think that, like, the DeepSeek models tend to be the most challenging as they have, like, the most novel architectural stuff going on, model after model. But every new model has something. Kimi K2 had. Oh, sorry, GLM-5.2 hadAli [00:14:53]: Sparse attention.Philip [00:14:54]: Yeah,Ali [00:14:54]: YeahPhilip [00:14:54]: the DSA.Ali [00:14:55]: Right. Which is brought from DeepSeek.Philip [00:14:57]: Yeah. AndAli [00:14:59]: So you can copy-paste then?Philip [00:15:01]: It kindAli [00:15:01]: I don't know how this works.Philip [00:15:02]: So, like we had to, like, build support for that into our runtime. And you're right, like it is really interesting the way that all of these open source labs borrow from each other. For example, like GLM-5.2 doesn't have vision. So something that, Haley, a guy on our team, if we could take a look at this, he, like, grafted the Kimi vision encoder onto GLM-5.2.Retrofitting Vision into GLM-5.2Ali [00:15:27]: We'll be training the projector.Philip [00:15:28]: Exactly. So if you think about, like, the encoder, there's the encoder, which is the part that looks at the image and turns it into latent information, and then there's the projector which likeAli [00:15:38]: You can say latent space. It's okay.Philip [00:15:41]: And then there's the projector that maps it onto, the model itself, and then there's the model weights. You don't wanna mess with the model weights because you run a chance of making the model dumber at something else for the purpose of giving it vision. So instead, Haley started with just a projector, which is only a handful of millions of parameters.Ali [00:16:02]: That would be, yeah.Philip [00:16:02]: Yeah.Ali [00:16:03]: Can you show the training one?Ali [00:16:04]: Like the way it groksPhilip [00:16:05]: YeahAli [00:16:06]: Very interesting.Philip [00:16:06]: And maybeAli [00:16:07]: That right therePhilip [00:16:07]: Maybe Ali, you should take it from here. You've got a betterAli [00:16:10]: Ooh, double the sandPhilip [00:16:11]: Understanding of this than I do.Ali [00:16:11]: Yeah. You can see, like, he. The way he trained this is really cool. At the beginning, he was training it using just like, “Here's a picture of a mountain. Can you describe what's in this mountain?” And that caused it just like the first, learning walls. Like here you can see this all we're trying to teach it is to translate the encoded. Like it's already taken the encoder from Kimi K. It's taken the image. It'Philip [00:16:31]: Yeah. FrozenAli [00:16:31]: FrozenPhilip [00:16:32]: With adapter.Ali [00:16:32]: Exactly.Philip [00:16:33]: Yeah.Ali [00:16:33]: So the brain is frozen and the eyes are frozen. It's just we're tryingPhilip [00:16:37]: AlignAli [00:16:38]: Interconnect between the eye and the brain, right? So the projector. And so you take the tokens and then he's like, “Oh, can you describe what's in this image?” And he's like, “Oh, it's a mountain,” or it's a person or it's a human, whatever the case is. But that didn't cause complete understanding. So he changed it such that every image was associated with a data set of questions. Like, does this image have a white male? Does this image have birds in the top corner? Does this image have a scientist in it? All of that stuff. And it would have to answer questions correctly. And using not just training on describing an image, but being able to answer question, another question, answer over time. Like you can see the grokking, which is like genuinely insane, that retrofitting vision into a large LLM can learn to that extent. And even for images that it doesn't perform well on, for instance, if you ask it a picture of like Stephen Hawking, “Who is this?” Maybe it doesn't get it, but it will say something like, “This is Albert Einstein.” Like it still understandsPhilip [00:17:25]: Close enoughAli [00:17:26]: That this is a scientist who is a man who has, some significant achievements, all that stuff. So that's like really cool.Philip [00:17:32]: Yeah. So, we've covered Hao Tian before, who the author of the LLaVA paper that did this, a while ago. And I think that's very foundational work for anyone who hasn't done vision work before.Ali [00:17:41]: Same with the CLIP and MetaCLIP, where you go from just captioning to building out questionsPhilip [00:17:47]: RightAli [00:17:47]: Off the image and how much better you can get performance.Philip [00:17:50]: Right. Right. Right. Yeah. But what's, what's so exciting about this is if you look at a model like this. Now, this is a little bit more of a research project. It's not. It got to 56% on MMLU Pro, I think. So not quite frontier. But if you're running this model, you haven't suffered any loss on your GLM-5.2 quality. If you don't have an image, it'll just behave exactly the way it used to. And ultimatelyAli [00:18:14]: Which in the inference code you literally do not include the other part, right?Philip [00:18:18]: Yeah. You would just skip the encoder if you don't have an image input.Ali [00:18:22]: Okay.Philip [00:18:22]: Just confirming.Philip [00:18:23]: YeahAli [00:18:23]: Does it affect a lot on the overall inference side? Like you're not adding much, you're adding a very small vision encoder. These are typically likePhilip [00:18:30]: They're super fineAli [00:18:31]: Less than a billion parameters, right?Philip [00:18:32]: Yeah. It's, - There's a little bit less standardization among vision encodersSwyx [00:18:37]: YeahPhilip [00:18:37]: So the support matrix can be a little bit, sparser. But overall, yeah, it's a pretty, it's a pretty minor component of the overall system. And ultimately what you get out of the system is all of a sudden you have Kimi Vision, GLM weights, and DeepSeek attention all in one model.Open Source Model Grafting and Franken-MergesPhilip [00:18:56]: And that's, I think, a lot of the power and beauty of open source, is that you can take all of these different components and combine them together into a system that's better than anyoneSwyx [00:19:05]: YeahPhilip [00:19:05]: Can be individually.Swyx [00:19:06]: People used to say that you would also do Franken-merges where you would take likePhilip [00:19:10]: YeahSwyx [00:19:10]: Layers from each model.Swyx [00:19:11]: Does anyone do that anymore?Ali [00:19:13]: Well, to your point previously when you were mentioning like, the work that goes into supporting a model when it first comes out, like GLM-5.2 or MiniMax M3 or whatever the case is. Sometimes you do have to like, you do have to switch out some things. Like, for instance, the MiniMax M3 head uses full attention, and with full attention you end up with this like insane bottleneck in spec dec ‘cause you're doing auto-regressive token generation for three tokens, and you're doing this like N squared over all of the tokens that are in your sequence. Your KV cache is like very large because it's not sparse, it's not top K. So we find it better to like, okay, we're gonna replace this, we're gonna replace this layer with a layer from another model that's using like GQA, for instance. And then just with the right training, you can get it to have the same acceptance rate. So it is very possible to retrofit layers from other models and very much needed. If a layer is like inefficient, the training just becomes the challenge, like how do you ensure that you train it properly? Which again to your earlier point is like the mesh between training and inference. As in like you need very good training in order to do fast inference. That's like, I feel like more and more becoming true.Swyx [00:20:21]: Yeah. Anything else on the support side when you say like get it to fully production ready?Loop Detection, Race Conditions, and Non-DeterminismPhilip [00:20:26]: Yeah. I think that there's also a question of just, we can test a model to a pretty extensive degree, but we're trying to get it out quickly and then you see a bunch of other people test it and you get interesting results. There was an issue with, GLM briefly where we had some like mode collapses where it would just output the same token over and over again for certain prompts on certain temperatures. Like once you expose an endpoint to the real world, there's going to be, so many more varieties of things given to it that you're able to, discover and patch things. So it's not just a, day zero process, it's then like for the first week, for the first month, if a model remains popular, like how do you both fix bugs and then continue to push the envelope on performance?Ali [00:21:21]: What do you mean you don't want your model outputting S?Swyx [00:21:24]: Is there loop detection on that stuff, by the way? It still happens like quite a lot, which is surprising.Ali [00:21:30]: We have like we, in our endpoint, like if a model was to output the same token like four plus times, we just cut the generation. We say like, “Oh, sorry, this-- Like try again,” or like we will reprocess the request. ‘Cause we know then, like if it, like if, yeah, it's four times the same token, it's probably collapsed.Swyx [00:21:45]: Yeah. Is there a way to opt out in case I really want that?Ali [00:21:48]: You want that?Ali [00:21:50]: I think there's a way that we have to handle it. I'm not exactly certain, but I feel like in certain models, like when they output something like you can imagine, like a table for instance, and so they want, they wanna draw like 12 dashes and 12 dashes. Yeah, I think there's a way for that to happen. I think we only do it on certain tokens. Like we exclude certain special characters.Swyx [00:22:07]: Yeah.Ali [00:22:07]: So we only do it on like certain like S is the most common almost. GLM-5.2Swyx [00:22:11]: OhAli [00:22:11]: And I think it was DSV 4 as well. Like you'd just have like looping issues where like you literallySwyx [00:22:17]: ItAli [00:22:17]: Just have like S.Swyx [00:22:18]: Yeah. Is there a special, something special about S? No, just randomlyAli [00:22:21]: It just seems to be the one token involved.Swyx [00:22:23]: Yeah. And it'Philip [00:22:24]: Is thereSwyx [00:22:24]: And it's only temperature 0Ali [00:22:27]: NoSwyx [00:22:27]: Even at other temperaturesAli [00:22:27]: Even at like 0.9 or whatever, it will still, it will still collapse.Swyx [00:22:30]: That's weird, right?Ali [00:22:30]: It's, it is an inference problem to be honest, like a software problem. Like oftentimes, the image you run will-- like NVIDIA will release an image for instance, and if we will upstream the changes from their latest TensorRT-LLM image into our stack, we'll find that it fixes it. Or oftentimes this will only happen in an inference engine that you're using like SGLang. But if you were to switch to vLLM, that isn't the case. So it seems to be like an extremely like deterministic software issue and not really a model issue. It's not like a weights problem. Like I'- we'll say like, “Oh, it's a problem with the quant. We did PTQ wrong,” right? But that isn't, that doesn't make sense because the same weights used with a different inference engine does not repeat the problem. And sometimes it's, the kernels that are being used in the backend have like these very subtle sometimes race conditions, where if you were to use this model hosted on one cluster, you will never get this problem.Swyx [00:23:19]: Oh my God.Ali [00:23:19]: But if you host it on a different cluster, you will. And the reason is the KV cache transfer from a node to node in that one cluster is using a slower interconnect than the node to node in another cluster. So that exposes the race, whereas in another cluster it doesn't. So then you end up just like, okay, this model is not gonna be hosted on this cluster. We're gonna host it on, another cluster because that cluster exposed that problem. But then it ends up with like, okay, is it the software? Is it the model weights or is it the hardware?Swyx [00:23:42]: There is a thing about this with temperature 0 still not being deterministic, right?Ali [00:23:46]: Right.Swyx [00:23:46]: Mostly because of hardware. Even at temperature 0 same model, you won't always get the same output.Swyx [00:23:52]: Even-- But I'm surprised by the race condition one because, I thought PyTorch was a graph that like guarantees that you at least, execute things in the right order.Ali [00:24:02]: Well, yeah, true. Like I'm not, I'm not saying that there is. Like well, you have things like PTL optimizations where like you can start a kernel before the end of the previous kernel, and that's like ‘cause you want to do that because there'sSwyx [00:24:12]: It's like pipeliningAli [00:24:12]: Expense. Exactly.Swyx [00:24:13]: Yeah.Ali [00:24:13]: But it'- But you don't do it cleanly. Like you overlap a little bit of the execution. No, it is very possible that the kernel itself, like that one block that is supposed to be running in this instance of time, that kernel itself has a race condition. For instance, like a missing barrier. Like often if you're designing a kernel and you want it to make it to be very fast, if you don't test it extensively, you'll, you'll have certain threads access data points from registers before they've been written to by other threadsSwyx [00:24:36]: YeahAli [00:24:36]: For example, because like your barrier is wrong or your synchronization was wrong. But yeah, like the testing itself is very difficult in those like, andSwyx [00:24:42]: And there's no like borrow checkerAli [00:24:45]: What does that mean?Swyx [00:24:46]: Like Rust. Like the. If you're trying to have like memory safety It sounds like a comparable problem.Ali [00:24:52]: Well, yes, but you're working in CUDA, right, NVIDIA GPUs. Like- You just need a higher level language like modular Maybe that's what modular is supposed to do. I don't know.Quantization Quality and Vendor FidelityVibhu [00:25:00]: How do you see keeping quality of the model? So you talked about all these steps of, okay, you gotta do quantization, train your own speculative decoderAli [00:25:07]: RightVibhu [00:25:07]: Run on different hardware. Looking at other model providers, okay, you kicked off a inference speed race on the consumer end. What goes into keeping quality the same across them, right? Sure, you can run benchmarksAli [00:25:22]: YeahVibhu [00:25:22]: But, like, how do you determine how much quantization are there standards? What goes intoPhilip [00:25:27]: There's a few things on quality. Most inference optimizations are lossless. KV caching, for example. You are just recomputing or preventing recomputing the same values. Speculation, of course, if a draft token is wrong, it gets rejected. The main lossy optimization is quantization. And that really comes down to, number one, data format, number two, which parts of the model you choose to quantize, which layers, and number three, like doing a lot of calibration on the quantized weights, to ensure that you're preserving all the outliers. There's other tricks that you can do, though. A big one is long context, ‘cause one thing you asked at, right at the beginning is, “Oh, what's gonna happen if I send a 200,000 token request in?” So with a long input sequence, you need to, store a lot more information. You need to process a lot more tokens. And so even if a model has a context of a certain length, you might, as an inference provider, choose to build an API with a shorter context length, and of course a full length one as well. Because if someone doesn't need the full million token context, for example, you can get them better performance. I don't know if that's exactly like quality of the model. The way that I think about quality is to what degree are we faithfully serving the original model? If you think of a golden implementation of a model that performs exactly the way the model is designed to perform, I think of quality as how close are we getting to that, 100% fidelity of the model.Philip [00:27:13]: You can also, of course, think about quality from the training side and how do you push yourself past 100%. But when I think about purely inference optimizations, it's getting faster while staying as close to that 100% fidelity mark as possible. And certainly our standard internally is that, like you should not be able to tell the difference between our API and a, official API. I think Kimi in particular does a good job of vendor benchmarking hereAli [00:27:41]: YesPhilip [00:27:41]: Where they haveAli [00:27:42]: They released an actual vendor benchmark.Philip [00:27:43]: Exactly, yeah.Ali [00:27:44]: ‘Cause they accused, some people, Amazon? There was some provider that was not doing very well on Kimi's benchmark.Philip [00:27:50]: Yeah.Philip [00:27:51]: So, with Reflect we probablyVibhu [00:27:52]: This was a long time ago, right?Philip [00:27:54]: No.Ali [00:27:54]: Yeah, like threeVibhu [00:27:55]: They alsoAli [00:27:55]: Four, five months agoVibhu [00:27:57]: This also happened with, I don't remember which model, but they pulled out quite a few, and then they started a whole chart about this. It might have beenPhilip [00:28:03]: Kimi Vendor Verifier.Ali [00:28:04]: Yeah.Philip [00:28:05]: Yeah.Ali [00:28:05]: Yeah, ‘cause you, ‘cause you'd be pissed, right? Like if you'Philip [00:28:07]: Yeah.Ali [00:28:07]: If like if I'm a consumer and I'm using like Amazon's endpoint for instance, and I've used Kimi and I'm like, “Oh my God, like this is bad,” I'm not gonna say, “Oh, Amazon quantized the model in a bad way.” I'm gonna say, “Oh, Kimi sucks.” Right?Philip [00:28:17]: Yeah.Ali [00:28:17]: So it seems like that makes sense.Philip [00:28:19]: Yeah, they care. They care.Vibhu [00:28:21]: Justifiably.Ali [00:28:21]: Yeah, justifiably.Vibhu [00:28:22]: This is probably a stupid question, but just checking, has anything improved from main quantization?Philip [00:28:28]: Yeah.Vibhu [00:28:28]: Like, is quantization always strictly worse?Ali [00:28:30]: Well technicallyVibhu [00:28:32]: NoAli [00:28:32]: It's a lossy. QuantizationPhilip [00:28:33]: YeahAli [00:28:33]: Is a lossy, it's a lossy implementation.Philip [00:28:36]: Speed improvesVibhu [00:28:36]: Speed improves.Ali [00:28:37]: It the number, likeVibhu [00:28:38]: No, I' always look for inverse scaling laws.Philip [00:28:40]: Yeah.Ali [00:28:40]: Yeah.Vibhu [00:28:40]: This is something I learned from Noam Brown, where like things that normally act in one direction sometimes do.Philip [00:28:45]: Well, technically when you run a benchmark, because these models are deterministic, sometimes your,Ali [00:28:52]: YeahPhilip [00:28:52]: NVFP4 quant is like, two basis points higher than yourAli [00:28:56]: No, it's noise. It's noise.Philip [00:28:57]: Yeah, exactly. I'm like, yeah, it's, it's within. That's why I always say within margin of error.Philip [00:29:01]: And I stopped saying that because everyone assumes that what is, well, within some margin of error, we're barely inside of that to the worst, so we're saying. But yeah, sometimes it's just like, gives you a higher output score. But like Ali said, that's noise. To my knowledge, you're not necessarily making the results better. You're just trying to, again, like keep your fidelity as close to 100% to the original model.Layer Selection, KL Divergence, and Better QuantizationAli [00:29:27]: There is, to your point, research that we did on MP. I don't know if you are able to pullPhilip [00:29:31]: YeahAli [00:29:32]: A tweet we did. One of our research interns, Joshua, I think it's a tweet on how we have 20% better quantized GLM-5.2 than NVIDIA. Essentially what we found throughout like this month research is, okay, quantization is a lossy. It's. You're compressing the data from, occupying 16 bits to occupying, four bits, for instance. And so you're losing some information, and you're trying to minimize that. And so when I say that I'm gonna quantize the model, my job becomes how do I find the layers that I can quantize, and how to find the layers to not. For instance, with image models, I don't quantize modulation layers, and I don't quantize out projections because those two are. Like out projection is what you see as the user. Modulation is what the model sees or understands. Right, exactly. And so to his paper, do you have the. It doesn't have the. Yeah. It's a long paper. I don't know if I can findVibhu [00:30:25]: If there's a part to search or it's probably in the thread.Ali [00:30:28]: It's probably in the thread.Vibhu [00:30:29]: Yeah.Ali [00:30:29]: But the long and the short is it is very possible that quantizing more of the model makes the results. Like if I have a model that I quantize layers one, five, and 10, and another model where I only quantize layers one and It is possible that the model in which I quantized more information is going to perform better because the quantization errors have canceled out. And so what Joshua showed in his mathematical proof where he had like a verifier in, is that you can predict which layers are going to have quantization errors that will cancel out with each other, and you choose to quantize those layers. And so the result of doing this mathematical quantization is you end up with a model that's 20% more quantized than another provider, so you get 20% more throughput of it because there's more layers than running an NVFP4, and your quality is better than that other quant because the layers that you chose to quantize have their errors cancel out, like one layer skewed to the right one layer skewed to the left, one layer skewed to the right. Your final logits distribution is more similar to the original distribution of the model, so you have better fidelity. And so the way we proved this was with KL divergence. So instead of just scoring on the benchmarks, we scored the KL divergence between the logit distribution of the quantized model and the logit distribution of the original full precision model, and we showed that with this technique we get. If your probability distribution on the logits which token it wants to select is more of the same as the original model, you're probably gonna end up staying true to the original model. So yeah, so it seems like previously before this, it seemed like the industry was, well, the more you quantize, the worse it's gonna be, ‘cause the more loss you introduce. That's not exactly, not necessarily true. So yeah, doesn't improve it, but can cancel out.Philip [00:31:57]: I think it might be this, but reminds me a good bit about pruning where you can prune off certain layers.Philip [00:32:03]: But very interesting. Didn't know this was a whole paper you guys put out.Ali [00:32:06]: It's. Fun fact, it was originally 72 pages, this paper, and then we decidedPhilip [00:32:11]: WowAli [00:32:11]: We can't tell. We couldn't release it. So it's now 45.Swyx [00:32:15]: Still 39 pages, so very substantive. We talked about evals and all these things and, like what's possible in terms of speedup? Like it's like probably like the numberInference Speedups and BenchmarkingSwyx [00:32:25]: Thing that people do wanna care about, and it's something that you wrote about in your post. Like official API is 70 tokens per second, and you push it up to 90. Is that like a normal thing?Philip [00:32:36]: So what's cool about working in inference, the reason that I think inference is going to be a useful place to do engineering for a long time, is that if you look at highly optimized domains like, say, finance, if you're in finance, you measure how much better you got in basis points. It's like, “Oh, I got five basis points better, like twentieth of 1% better,” that's huge news because everything is so optimized. When we publish optimizations, it's 20%, it's 100% it's 200%. So there's still probably like a lot further to go, honestly. Like you'll, you'll know that inference is pretty much solved when researchers start publishing about how they got 1% faster at something.Swyx [00:33:19]: Which by the way, because I am from the finance background, in the ‘70s, that was the margin at the time. When you did quantitative finance research, you would findAli [00:33:27]: And like 20%, tens of percent.Swyx [00:33:29]: That's. Yes.Philip [00:33:29]: Yeah.Swyx [00:33:30]: And now it'Philip [00:33:31]: Tiny fractionsSwyx [00:33:32]: For those people interested, look up Andrew Lo's paper. He had a really interesting illustration of quant, stat arb, distribution, narrowing down from like those kinds of 20% differences in the ‘70s, down to nothing today, which is very cool.Philip [00:33:48]: Exactly, and we're at the beginning of the same type of thing. Now benchmarking is hard. I think anyone will tell you that, and benchmarking provider speeds is hard because there's so many variables that go into it. What hardware are you using? How much load do you have on the system? What's the exact nature of the prompts and input and output sequence lengths? All that stuff. But overall, when you start stacking these improvements, you're looking at multiples. You can look at it. The most common form, of course, is TPS, tokens per second, which is bad naming by us in the industry, ‘cause there's two tokens per second. There's tokens per second, the throughput number, and the latency number.Ali [00:34:31]: TTMT, yeah.Philip [00:34:32]: Like total tokens per second out of the, out of the GPU as a throughput number. Most people only care about tokens per second as the latency number, which we should call ITL, intertoken latency, but we don't.Philip [00:34:44]: Anyway, so you can imagine a standard API without many optimizations for a 1 trillion parameter model operating somewhere in the 30 to 50 tokens per second range for reasonable traffic profile. And we generally see the goal of, pushing to 10X that. But, not necessarily day zero, but by stacking enough optimizations, if you have, say like four optimizations, each of which doubles performance. Or sorry, three optimizations, each of which doubles performance, then you stack that up, that's an 8X gain. That's the order of magnitude that we're working with in this space. We're trying to make things substantially faster, not just go from like 70 to 90.Swyx [00:35:38]: Are you saying you've. You have done that?Philip [00:35:40]: So let's say you have as a reasonable baseline, 30 or 40 tokens per second. You can achieve 10X that. So like on GLM-5.2, if you run it unquantized, perhaps on H100s even, and you're just using an off-the-shelf inference engine with no particular optimizations, no speculator, nothing extra around like KV routing, no disaggregation, you're, you're probably, yeah, looking at that like 30 to 40. You think that's like a reasonable baseline?Swyx [00:36:12]: Right. Right.Philip [00:36:12]: To get to something like 10X, there's a lot of trade-offs that you're making. If we're running at more like a 300, 400 tokens per second range, you are using the best hardware possible. You have a optimized speculator. You have done all of your quantization work. You are Seeing a pretty high cache hit rate. You are running with a reasonably small batch size and a parallelism configuration that is tuned for latency versus throughput, but it is possible. So the spreads that you see if you, like, go on artificial analysis or you go on OpenRouter and you look at, the worst provider to the best provider, oftentimes can hit that range. 10X is of course very aggressive. It's oftentimes maybe more of a four to six times improvement. But that's the performance that makes us really excited, is when we can get these huge gains, not just go from 70 to 90 tokens.Stacking Optimizations: NVFP4, Speculation, and DisaggregationAli [00:37:19]: It's also, like, hardware dependent. Like, ifPhilip [00:37:20]: YeahAli [00:37:20]: If you have a thing where you're serving it on just, like, a node of H100s and then you throw, like, you shard the model across, like, four nodes of B200s. Like, you can definitely increase the speed with just throwing more hardware at it. Like, normalizing for the same exact hardware and the same number of GPUs.Philip [00:37:35]: Yeah. Then you're looking at, like, a two to 4X improvementAli [00:37:38]: Right. RightPhilip [00:37:38]: Depending on the inference optimizations. So yeah, it's. Some of it's, what's the call, and some of it's who's the driver.Vibhu [00:37:46]: If you break down the two to 4X, say the example is run GLM-5.2Ali [00:37:51]: YeahVibhu [00:37:51]: On B200sAli [00:37:53]: YeahVibhu [00:37:53]: Single node, right? What's, like, the cost trade-off for effort to get, like, the last bit of juice out versus what should people just think of, right?Ali [00:38:01]: Spectre quantization. Yeah.Vibhu [00:38:03]: Spectre quantization.Ali [00:38:04]: That's, that's, that's like 95%. LikeVibhu [00:38:06]: And how far does that get you? And how easy is that for the average person to do? So say right I wanna throw the weights of GLM-5.2 on a node of B200s, how easy is it to find speculative decoder- decoder model or already quantized model? How much work goes into it?Philip [00:38:23]: If you're doing it up front, it's quite a lot of work. If you're doing it today, there's going to be people who have published things that you can just, you can just grab some NVFP4 weights. You can grab a speculator. Yeah, if we're thinking about, like, what are the 2Xs we're stacking, going from, BF16 to NVFP4 is, it's not quite a 2X, right? It's like. I think it's about, like, 30 to 40%, from 16 to 8, and then another 30 to 40% multiplied from, 8 to 4. So that doesn't quite get you a 2X, but, like, roughly a 2X. Speculator, roughly a 2X. Disagg on top of that if you're able to get enough hardware and put enough traffic through it, another roughly a 2X. And then you add in some, double-digit percent increase from having just a better runtime with, the latest kernels and stuff behind it. And that's how it stacks up.Ali [00:39:21]: YeahPhilip [00:39:21]: So building each of those, like, building the, quantized weights is, for someone who really knows what they're doing, hours to days of work. Building the speculator, again, like, hours to days of work. And the, disagg setup, hours to days. Well okay, but like once you haveAli [00:39:39]: Once set up. Once set up. YeahPhilip [00:39:40]: Yeah, getting disagg working for the first time, I'm saying, of course, is very difficult.Philip [00:39:44]: The marginal implementationAli [00:39:48]: Like, if you're just grabbing, like if you are a person, like just a normal consumer who has access to, like, a node of B200s and you're wondering, “How can I just host it myself?” You don't need to quantize the model yourself. There's always gonna be, like, an open source quantized checkpoint. NVIDIA's gonna push one out if no one else does. You. Usually, the providers will have their own spec dec that they've trained as well. You don't need to train your own spec dec. You can just use that as well.Philip [00:40:09]: Yeah. Like, GLM-5.2 has its own MTP.Ali [00:40:13]: Right. Right.Vibhu [00:40:14]: What's multi token prediction?Philip [00:40:15]: Yes.Ali [00:40:16]: I'm justVibhu [00:40:16]: Can you explain that?Ali [00:40:16]: I'm just an expert.Ali [00:40:18]: I can do it for you in case I get it wrong?Vibhu [00:40:20]: No.Vibhu [00:40:21]: Yeah, you should correct if we're wrong, but their multi-token prediction can be used for self-speculative decoding.Ali [00:40:27]: I'm not sure. I'm not gonna correct that.Vibhu [00:40:28]: Okay. I'm semi-confident in thatAli [00:40:30]: Okay. YeahVibhu [00:40:30]: But someone can check. But it's useful to paint the story of, okay, not just the average person, but say a company wants to switch from serverless inference I wanna throw this up on. I wanna rent some GPUs, throw it up. These are the steps you take to do significantly faster than just put it behind vLLM.Ali [00:40:48]: Right.Vibhu [00:40:49]: I was waiting for a mention of Dynamo.Vibhu [00:40:51]: I feel like, that's supposed to be the baseline that you measure against.Dynamo, KV Routing, and Disaggregation ToolkitsPhilip [00:40:55]: I would think of Dynamo as less of a box system and more of a toolkit for building with. So when we talk about doing aware routing, when we talk about doing KV offloading, when we talk about doing, PD disaggregation, Dynamo fundamentally is. By the way, Dynamo is an open source library from NVIDIA.Ali [00:41:17]: We've done a pod with KylePhilip [00:41:18]: OkayAli [00:41:19]: Kyle Cranin.Philip [00:41:19]: Cool. So then your listeners know then that it supports all the different inference frameworks. And it is multi hardware, which is interesting.Ali [00:41:28]: But it's just a router, it's not like an optimizer layer.Philip [00:41:30]: Yeah. All it does, like, what Dynamo is good at, it is a library for moving information around your cluster, around your hardware. So if you have, KV cache on one place and you need it to be somewhere else, Dynamo coordinates NIXL for you to move that around.Philip [00:41:49]: That doesn't mean that, like, out of the box, you just say, “Pip install Dynamo,” and then you get, like, a massive performance speed up. It's more of a developer toolkit.Ali [00:42:01]: Yeah. I would have said it would. It comes with a set of defaults that you can then swap out.Philip [00:42:06]: It does. If the industry at large, I think, was, like, rolling out all of these deployments, standard, then I think it would be, like, a credible baseline. But, we've got to, we've got to benchmark against, like, what we're seeing in the wild.Speculative Decoding Methods: Medusa, EAGLE, n-Gram, and Spec-SpecVibhu [00:42:23]: I did wanna talk a little bit more about PD disagg, because that is probably, like, number three after quantized and speculative decoding. In your book though, I was just gonna pull out the book.Philip [00:42:31]: Yeah.Vibhu [00:42:32]: Like section 522 on Medusa, 523 on EAGLEPhilip [00:42:35]: YeahVibhu [00:42:36]: 524 on gram.Philip [00:42:37]: It's 55, would be disaggregationAli [00:42:42]: Yeah. Well, no, I just wanted to dwell a little bitPhilip [00:42:44]: YeahAli [00:42:44]: The other. Like, so what do you choose to include? What do you choose to not to include? Because there was all these other techniques.Philip [00:42:51]: Yeah.Ali [00:42:51]: Are these still relevant? Because I think they came out, like, a year and a half ago maybe.Vibhu [00:42:55]: Medusa is quite old.Philip [00:42:56]: Yeah, Medusa's old.Ali [00:42:58]: It was old.Vibhu [00:42:58]: But is it in the book as a good, here'sPhilip [00:43:01]: BaselineVibhu [00:43:01]: Baseline vanilla understand it?Philip [00:43:02]: Like you should know this.Vibhu [00:43:03]: Like I read the paper, I'm like, “ it makes so much sense.”Philip [00:43:05]: Yeah.Philip [00:43:05]: So with the book, I had a couple goals. One was to give people just a working vocabulary for the space as a whole, and the other was to give them some intuition about how each of these techniques works. As I mentioned in my AI Engineer talk, which is the first public addendum to this, the speculation space has moved much faster than everything else. So yeah, even at the time that I wrote the book Medusa, I very much included as a way for people to understand how the space evolved rather than what the most modern technique is. And now of course, there's DFlash, dSpark. There's, there's newer techniques even than EAGLE, although EAGLE is still very commonly used.Ali [00:43:51]: SpecSpecta.Philip [00:43:52]: Yes. Speculative decoding.Vibhu [00:43:54]: What canAli [00:43:56]: Oh, it's a paper by Tri Dao and it's like, it's doing speculative decodingVibhu [00:44:00]: HuhAli [00:44:01]: For the speculative decoder.Philip [00:44:02]: Oh, in spec- oh my God.Ali [00:44:02]: It's literally just an another. It's like, yeah, that's the most simple way to explain it, and it seems like he got trivial speed ups there. But it seems that the complexity with training, it's almost like in our mind at least, it's almost as complex as training GANs. Like it's like a very delicate balance and oftentimes you, it's just but yeah, it's literally speculative decoding on speculative decoding.Vibhu [00:44:21]: Speculative.Ali [00:44:22]: Yeah. We saw this paper.Vibhu [00:44:24]: It's interesting, right?Ali [00:44:24]: Yeah.Vibhu [00:44:24]: I wouldn't even expect it to be very particular to train, I wouldAli [00:44:29]: Right.Vibhu [00:44:29]: The naive part of me is like, okay, train speculative decoder.Ali [00:44:32]: But like, and it makes sense, like the whole idea of speculative decoding is you. It's like, it's like almost like the iPhone auto predict version but for a normal model, right? Like you're just, you're just, generating three tokens and you're like, okay, I'll do prefill on them. And so you save those three turns for your original model. Now your speculative decoder is doing three turns of auto regression, so why not just have an even smaller model?Ali [00:44:53]: The other question there is what are the size of speculators? So say forPhilip [00:44:58]: Right. It's like a billion parameters.Ali [00:45:01]: Like for MiniMax, it's. Yeah. It's like one layer. It's like one 60th of the original model usually.Philip [00:45:06]: Yeah. I think we should do a paper when we get back to the office.Philip [00:45:10]: SpeculativeAli [00:45:11]: SpeculativePhilip [00:45:11]: Decoding.Ali [00:45:13]: No, it's, it does seem like how, when do you stop? But then it also seems like if you're able to train spec-spec decode for instance, right? Like if you're able to have a small model that is accurately predicts what the intermediate speculator is gonna predict, that is able to predict what the original target model's gonna predict, then why not just use that smallest model directly, right?Vibhu [00:45:34]: Yeah. This isAli [00:45:35]: Like it seems likeVibhu [00:45:35]: Adjacent to the routing problem.Ali [00:45:36]: Right.Vibhu [00:45:36]: Yeah.Ali [00:45:36]: Right.Philip [00:45:37]: The thing with speculators is one of the practical constraints on using them is that you do have to run a small model on the same hardware that you're running the big model on. There is a orchestration and resource competition problem inherent in that, and that is one of the constraints on speculation in general, is that draft tokens cost resources to create and cost software complexity to manage. And so if you have like infinitely recursive speculators, you add in quite a bit of that complexity on the actual implementation within the inference engine as well, not just in the training process.Vibhu [00:46:17]: I was gonna say, I would wonder if you could do similar, like distillation and pruning of, it's the same thing, it's just a model. Can we not just distill a lot of the weights, quantize the speculator, out of my domain? The question that also comes up is, this is all for big server workloads, right? How much of this applies to, say I have this MacBook, I wanna run Gemma really efficiently. Similar problems, not the same?Local AI vs. Data Center InferencePhilip [00:46:45]: Pretty different. I talked to Selo, about this on his podcast a couple weeks ago. The difference between inference engineering for the data center and for production workloads versus inference engineering for local AI, is that we start with fundamentally like different constraints and different goals. With local AI, it's how do I fit this model onto my hardware and then make it less dumb? And with data center influence, it's how do I load this model and then make it less slow? And we care about less dumb, and they care about less slow. But the local AI inference engineering ecosystem, I think has a lot for us to learn from in the data center space. They are experts in various forms of quantization, including dynamic quantization that we just don't touch, in the pruning, in the distillation, in the, layer removal. There'Ali [00:47:42]: Layer removal matters less.Philip [00:47:43]: Yeah. There'Ali [00:47:44]: No one loves pruning really.Philip [00:47:45]: Yeah. Well, but the, but they doVibhu [00:47:46]: Which is surprising, right? But that's, that's a whole different thingPhilip [00:47:48]: Just to fit something on the laptop.Ali [00:47:50]: Right.Philip [00:47:50]: So yeah, it's a, it's an interesting, it's an interesting space. Not necessarily that like their techniques make sense for us to do in the data center, because we have different resources and different goals, but more that the process as well as the openness of that field is something to, admire.Ali [00:48:12]: Yeah. Like to your point, like, certain optimizations that would. Like for instance, Turbo Quantum Sharper, like it made such huge hype on that and we did like a whole deep dive on Twitter and like said, what is it? How does it work? Why is it good or not? And it took off and it was implemented on local devices because your memory bandwidth is so slow on like a MacBook, for instance. But try putting the same thing on like an NVIDIA GPU on a B200 Turbo quant would not be. Like, it would not be used. Like, NVIDIA - Like, NVIDIA made it clear that this is not a good optimization, and we've seen it firsthand where the overhead of doing dequantization, quantization of, in the kernel itself with turbo quant kernel, each end is much slower than the time that you save from doing the bandwidth. ‘Cause on the B200s, you have like 3.5 terabytes per second. You don't need decrease the storage that much. You don't need to do, FP4 KV cache. You don't need to use a requant. There's, there's, there's better optimizations to be made. But on Edge devices, it's extremely important, it's extremely useful. So, seems to be, like, different optimizations there, but then they're all uniquely combined with like all you wanna quantize the model, you wanna do speculative decoding, like certain common prefixes with bothPhilip [00:49:18]: Principles.Ali [00:49:19]: Yeah, exactly. Exactly. Exactly.Philip [00:49:20]: They also do a lot of work on, model parallelism, especially over, heterogeneous topology, where you have, some sparks and they are wired together with, Ethernet, DGX sparks.Ali [00:49:35]: Yeah, this is the Exo Labs guys.Philip [00:49:36]: Yeah. You have, a nu

Et si la réussite n'était pas seulement une question de talent, mais aussi de persévérance, de chance et des choix que l'on fait chaque jour ? Pourquoi a-t-on parfois l'impression de ne pas être à la hauteur ? Comment continuer à avancer quand le doute s'installe ? Et qu'est-ce qui change vraiment lorsqu'on atteint enfin ses objectifs ?Pour explorer ces questions, je retrouve trois humoristes aux parcours très différents. Avec Kev Adams, on parle d'ambition, de santé mentale et de cette quête qui ne s'arrête jamais, même après le succès. Avec Nordine Ganso, on s'interroge sur le prix de l'ambition, les sacrifices qu'elle demande et la solitude qui peut parfois l'accompagner. Enfin, avec Paul de Saint-Sernin, on revient sur le temps qu'il faut pour construire une carrière et sur les choix qui façonnent une vie.Au fil de ces conversations, on comprend que la réussite n'efface ni les doutes ni les remises en question. Elle invite surtout à redéfinir ce qui compte vraiment et à construire un chemin qui nous ressemble.Je vous souhaite une très bonne écoute !——Pour découvrir les coulisses du podcast : https://www.instagram.com/inpowerpodcast/Pour suivre Kev Adams : https://www.instagram.com/kevadams/Pour suivre Nordine Ganso sur les réseaux : https://www.instagram.com/nordine.ganso/Pour retrouver Paul de Saint-Sernin sur les réseaux : https://www.instagram.com/pauldesaintsernin/Et pour suivre mes aventures au quotidien : https://www.instagram.com/louiseaubery/Chapitrage :00:00:00 - Intro00:00:22 - Paul de St Sernin00:11:05 - Kev Adams00:13:30 - Nordine Ganso00:20:38 - Kev Adams Hébergé par Acast. Visitez acast.com/privacy pour plus d'informations.

Amélie raconte un tour du monde en famille et une soirée dans un marché de nuit en Thaïlande, avec son mari Dylan et leur fils Aaron. Habitué à jouer à cache-cache tout seul, son fils disparaît dans la foule sans un mot, le temps qu'elle tourne le dos. Écoutez ce récit pour découvrir comment cette recherche affolée s'est terminée.-----------Si l'épisode vous a plu, laissez-moi une note 5 ⭐️ou un commentaire sur Apple Podcasts ou Spotify

L'Europe voulait un champion souverain. Elle obtient un partenariat avec Microsoft. Le deal Mistral-Microsoft divise : capitulation pour les uns, réalisme économique pour les autres. Mais derrière les déclarations officielles, une réalité s'impose — avec 4 milliards d'un côté et 190 milliards de l'autre, le choix n'était peut-être plus un choix.===================⏱️ DANS CET ÉPISODE :===================0:00 — Intro0:42 — [Sponsor] : Qonto, gérez votre facturation électronique sans effort !1:56 — Microsoft et Mistral signent : capitulation ou réalisme ?4:35 — Qui gagne vraiment dans ce deal ?8:46 — L'État français : présent mais trop pauvre pour peser14:08 — Data centers : ce que les discours ne disent pas15:22 — Brad Smith francophile : l'homme qui a fait pencher la balance16:36 — Partenariat tactique ou dépendance consentie ?=============

Logan Councilmember and Cache Water District board member Jeannie Simmons -- Could Utah's population projections be wrong? 

Clue one was a gentle lob right over the plate. Clue two came in and completely rocked our confidence. Caden, Christian, and Chad dig into Bank of Utah's second clue for the "Where's the Wallet" Cash Quest, and things get weird fast — is it about locomotives, automotives, or just plain old motives? Is Box Elder County still in play, or did we all fall for the misdirection the clue literally warned us about?We revisit the Box Elder theory (sweet sap, unsettled territory, Promontory Point train ties), float a Weber County curveball, and yes, someone brings up Cache County purely because it lets us say "Cache my Cache Quest" out loud. Buckle up.Got a theory better than ours? Send it our way — happy to read it on the show, or keep you anonymous if you'd rather just quietly be smarter than us.Leave us a review! Every episode we pick a favorite and send that reviewer $25. Make us laugh, make us think, or just tell us we're wrong about Box Elder.In This Episode:Caden's 5-year-best solve streak gets a shoutout (thanks, dude)Six-legged vs. eight-legged, and Highway 8 vs. Highway 6"We're cashing up" — an instant classic, we don't make the rulesLocomotive → automotive → literal motive, a philosophical journeyBox Elder trees have sap you can tap, and now you know tooThe Cache County pitch that may or may not just be brand loyaltyA Utah trivia bonus round, because five words really can be "Under the Banner of Heaven"

The break comes to an end (of it ever happened): ranking roster changes, free agents stocks, previewing the second season, and more in this episode of Confirmed!➡️ Follow us for updates: https://twitter.com/HLTVconfirmed

Extrait qui compile plusieurs passages avec différentes invitées, où nous évoquons le rapport au corps, à la séduction, les critères physiques, le mal que ça nous a fait d'être cachée, rejetée parfois, et la joie de réussir à séduire ceux/celles qui nous semblaient hors d'atteinte (ce que je nomme "phase palmarès"). Parce que oui, on a le droit d'être des personnes désirantes et désirables, même quand on ne correspond pas aux normes de beauté capitalistes occidentales.Ce qui fait un bon lien avec l'épisode avec Morgane Tocco, docteure en anthropologie sociale, diplômée de l'EHESS, autrice du livre "Moi aussi je te regarde - Regards de femmes sur corps d'hommes" aux éditions du Détour. https://singlejungle.lepodcast.fr/ep-point-135-morgane-tocco-moi-aussi-je-te-regarde-regards-de-femmes-sur-corps-dhommes Et avec cet épisode du podcast du magazine Elle, "il était une (première) fois" auquel j'ai participé :https://podcasts.elle.fr/elle-il-etait-une-premiere-fois/202403140652-sex-therapy"Sur une application de rencontre, Louisa fait défiler les profils. Elle tombe sur celui d'un homme tellement beau qu'elle se demande ce qu'il peut bien faire sur ce genre de plateforme. Ils commencent à discuter et elle s'interroge alors sur ce qu'il peut bien lui trouver. Un premier rendez-vous est décidé. Dans les rues, toutes les filles se retournent sur son passage, mais c'est elle qu'il a choisie. Leur première nuit d'amour est torride, inoubliable. En suivront de nombreuses autres qui, à chaque fois, lui feront l'effet d'un shot de confiance en soi."

This week we welcome back the fan favorite 1UpMe contest! This month? Pimp Your Cache! Tell us about the best cache you've hidden! Good placement? Good container? Let's see who takes the crown! Thanks to Big B'a Cache Supplies. Call in and try and win!

C'est la fin des quarts de finale de cette Coupe du Monde avec Norvège-Angleterre et Argentine-Suisse. L'Argentin vous a-t-elle convaincu sur ses deux derniers matches ? Messi est-il l'arbre qui cache la forêt ? Que penser des perforamnces de Lautaro Martinez et Julian Alvarez ? La Suisse a-t-elle tout à gagner ? L'Angleterre est-elle prête pour son quart de finale ? La mentalité des Three Lions est-elle différente que sur les autres tournois ? Quel est le principal mérite de Tuchel ? Haaland peut-il faire douter l'Angleterre ? La Norvège peut-elle créer l'exploit ?Ce podcast est hébergé par Podcastics, la plateforme pour créer et diffuser votre podcast facilement.

REDIFF - Dans ce nouvel épisode de "Symptômes", le médecin généraliste Érik Bernard nous raconte l'histoire d'une jeune femme de 28 ans, venue consulter pour une hypertension artérielle assez banale. En bonne santé apparente et sans symptômes alarmants, la patiente ne s'attendait pas à ce que sa visite de routine révèle une situation bien plus complexe... Retrouvez chaque mois, un nouvel épisode inédit de "Symptômes", ainsi qu'un bonus la semaine suivante.Hébergé par Audiomeans. Visitez audiomeans.fr/politique-de-confidentialite pour plus d'informations.

On today's show, we learn that CACHE's new Take Action grant program is designed to provide behind-the-scenes assistance to artists and creatives. We also get on a bike for the newest tour of Bentonville. Plus, Michael Tilley from Talk Business & Politics gives us a week-in-review.

Control what you're billed on. Tokens are the currency of AI, and how you design your app determines how many you spend. Compress conversation history instead of resending it raw, cap output tokens with matching prompt instructions, and cache static context so each reuse costs a fraction of the first request. Route each prompt to the right model by complexity, or set Model Router in Microsoft Foundry to handle that automatically — balanced, quality, or cost mode. Then optimize the whole stack. Run Agent Optimizer to test your prompt, model, and tool configurations together and surface better setups. Use Toolbox to dynamically select only the tools each request needs and cut input token overhead by 90%.  April Gittens, Microsoft Principal Cloud Advocate, joins Jeremy Chapman, Microsoft 365 Director, to share how to seize control of AI token spend through smarter app design.  ► QUICK LINKS:  00:00 - Tokenomics foundation 01:06 - Token cost basics 02:11 - Context Window creep 03:46 - Reduce unnecessary tokens 04:29 - Trim context costs 05:21 - Cap output tokens 06:27 - Cache for savings 07:54 - Model cost tradeoffs 09:15 - Model Router 09:42 - Toolbox in Microsoft Foundry Toolbox 11:18 - Agent Optimizer 12:44 - Other cost drivers 13:57 - Wrap up ► Link References Check out the tools in Microsoft Foundry at https://ai.azure.com For more about managing AI costs go to https://aka.ms/FoundryTokenomics ► Unfamiliar with Microsoft Mechanics? As Microsoft's official video series for IT, you can watch and share valuable content and demos of current and upcoming tech from the people who build it at Microsoft. • Subscribe to our YouTube: https://www.youtube.com/c/MicrosoftMechanicsSeries • Talk with other IT Pros, join us on the Microsoft Tech Community: https://techcommunity.microsoft.com/t5/microsoft-mechanics-blog/bg-p/MicrosoftMechanicsBlog • Watch or listen from anywhere, subscribe to our podcast: https://microsoftmechanics.libsyn.com/podcast ► Keep getting this insider knowledge, join us on social: • Follow us on Twitter: https://twitter.com/MSFTMechanics • Share knowledge on LinkedIn: https://www.linkedin.com/company/microsoft-mechanics/ • Enjoy us on Instagram: https://www.instagram.com/msftmechanics/ • Loosen up with us on TikTok: https://www.tiktok.com/@msftmechanics

REDIFF - Vanessa Lagesse avait 32 ans. C'était l'héritière d'une famille appartenant à la bonne société de l'Île Maurice. A la fin de l'hiver 2001, son corps est découvert dans son luxueux bungalow. Massacrée avec une violence inouïe. Un crime haineux, passionnel, acharné. La police de l'île va faire défiler les témoins disparates, parfois douteux, avant de s'arrêter sur un homme d'affaires en vue, Bernard Maigrot. Cet homme marié était l'amant discret de la victime. Leur liaison commençait à faire jaser. Il voulait la quitter, elle s'accrochait. Retrouvez tous les jours en podcast le décryptage d'un faits divers, d'un crime ou d'une énigme judiciaire par Jean-Alphonse Richard, entouré de spécialistes, et de témoins d'affaires criminelles. Cet été, retrouvez tous les jours en podcast le décryptage d'un faits divers, d'un crime ou d'une énigme judiciaire par Jean-Alphonse Richard, entouré de spécialistes, et de témoins d'affaires criminelles.Hébergé par Audiomeans. Visitez audiomeans.fr/politique-de-confidentialite pour plus d'informations.

An anonymous researcher has dropped a giant cache of zero-day exploits, a sensitive DHS network got hacked, the US Supreme Court restricts geofence warrants, and security firm Huntress has denied accusations of a malicious insider. Show notes Risky Bulletin: Researcher drops giant cache of zero-days

Send us Fan MailComedian Paul Bragin is back in the studio for his lucky-number-seven appearance, and the wheels come off almost immediately. This week on The Days Grimm Podcast, Brian, Thomas, and producer Corey turn a simple "deaths of the month" check-in into a no-topic-is-safe deep dive that somehow connects a horse-racing legend, a paramount principle of Jewish law, and a dictator's 1987 plan to breed cows the size of dogs. If you like your comedy podcast unscripted, unfiltered, and wildly off the rails, you found the right episode.From there it only escalates: the crew debates whether a pig heart transplant counts as kosher (turns out there's a real answer), digs into the melanin-peptide tanning trend and the surprising cancer research behind it, and settles into a heated greatest-director argument over Christopher Nolan's upcoming Odyssey. You'll get the real backstory of how Henry Ford "borrowed" the assembly line from Toyota, a field guide to Amish Rumspringa, and a passionate case for why Waffle House needs to come to Evansville — fights and all.Recorded in the Days Grimm studio in Evansville, Indiana, this episode closes with Paul's report from his longest-ever clean stand-up set, a tour through the chaos of the Evansville Watch local-news feed, and a stack of local plugs you'll actually want to hit. Grab a Four Roses (hopefully one day a sponsor), subscribe, and settle in.If this one made you laugh, do us a favor: smash that Subscribe button, ring the bell so you never miss a Tuesday drop, and drop a comment below telling us your hottest take from the episode. New episodes of The Days Grimm Podcast every week — and hey, Four Roses, our DMs are open.

REDIFF - Pendant l'émission, Sylvie Tellier a toujours un carnet avec elle. Mais que cache-t-il vraiment ? Fous rires, réponses inattendues, nouvelles rencontres, redécouvrez cet été les meilleurs moments de cette saison 2025-2026 !Hébergé par Audiomeans. Visitez audiomeans.fr/politique-de-confidentialite pour plus d'informations.

INÉDIT - Dans ce nouvel épisode de "Symptômes", le kinésithérapeute Antoine Couly raconte l'histoire d'une étudiante de 19 ans, née avec un lourd handicap, qui tente malgré tout de préserver son autonomie et de reprendre confiance en son corps. Au fil des séances, une douleur étrange et répétée aux poignets vient bouleverser le suivi. Face à des symptômes dispersés, longtemps banalisés, le praticien comprend peu à peu qu'un détail en apparence secondaire pourrait bien relier toutes les pièces du puzzle. Retrouvez chaque mois un nouvel épisode inédit de "Symptômes", ainsi qu'un bonus la semaine suivante.Hébergé par Audiomeans. Visitez audiomeans.fr/politique-de-confidentialite pour plus d'informations.

Send us Fan MailA motorcycle can cover distance fast, but that's not the point of this ride. We're chasing something quieter: a clear head, a lighter heart, and the kind of focus you only find when the road starts twisting and the mountains start towering. From Loveland, Colorado, I take you on a two-day, 350-mile Colorado Rockies loop designed for riders who love traffic-free roads, remote valleys, and the feeling of being small in the best way.We climb through the canyons and along the Cache la Poudre River, stop for a big breakfast at a local landmark, and roll into the wide-open stillness of Walden and North Park where fuel range and wildlife awareness actually matter. Then we cross Rabbit Ears Pass into Steamboat Springs, dig into the history behind the famous F M Light and Sons signs, and settle into pure Americana at the Rabbit Ears Motel before an easy evening by the Yampa River.Day two turns epic as we aim for Rocky Mountain National Park and Trail Ridge Road, the highest continuous paved road in America. We cruise the Kawuneeche Valley past willow marshes and the Colorado River headwaters, climb above tree line toward Milner Pass, and soak in views that stretch all the way toward Wyoming. We wrap with a coffee stop at the Stanley Hotel in Estes Park, sneak onto a lesser-known back road, and ride home with one simple reminder: we ride not to escape life, but to make sure life doesn't escape us. If this ride gave you a calmer mind, subscribe, share it with a riding buddy, and leave a review so more riders can find the peaceful route.Thank you Episode Sponsor - Viking Bags:   https://www.vikingbags.com/Viking Bags BMW R1250 GS Adventure Touring Hard Side Cases:https://www.vikingbags.com/collections/bmw-r-1250-gs-adventure-touring-hard-side-casesWant to Support the Podcast?Become a Member: https://www.buzzsprout.com/2126578/supporters/newBuy Ron a Coffee: https://buymeacoffee.com/peacelovemotoGear Up at the Shop: https://peacelovemotostore.com

Drama on a ThursdayFirst,  a look at this day in History.Then, Jeff Regan Investigator starring Frank Graham with Frank Nelson,  originally broadcast  June 25, 1950, 76 years ago, No Sad Clowns For Me. A circus story.  Mr. Crackly wants to find a man named Bliss. Followed by Suspense, originally broadcast June 25, 1961, 65 years ago, Call Me at Half Past starring Elspeth Eric and Bernie Grant.  A man is trapped in a hotel room by his insane wife, who is determined to kill him.Then, Let George Do It starring Bob Bailey,  originally broadcast  June 25, 1951, 75 years ago, The Man From Jaune Cache. Amnesia, man overboard and a prodigal brother-in-law arriving from South America. Let George do this one!Followed by I Was a Communist for the FBI starring Dana Andrews, originally broadcast  June 25, 1952, 74 years ago,  A Riot Made to Order.  Cvetic uses a sprinkler system to foil the plans of the Party to cause a riot and create sympathy for the Communists.Finally, Claudia, originally broadcast June 25, 1948, 78 years ago, The Mayor Discusses a New School.   A meeting with Mayor Reynolds. Kathryn Bard and Paul Crabtree star.Thanks to Bill B for supporting our podcast by using the Buy Me a Coffee function at http://classicradio.streamFind the Family Fallout Shelter Booklet Here: https://www.survivorlibrary.com/library/the_family_fallout_shelter_1959.pdfhttps://wardomatic.blogspot.com/2006/11/fallout-shelter-handbook-1962.html

It's a very sad episode this week on Overtime on Inferno, because Jack didn't back karrigans move to Falcons, he has identified himself as a fraud who doesn't understand Counter-Strike. In fact, this brought him so much shame that he has decided to leave the podcast so he can restore honour to his bloodline.Oh and by the way, we also talk about how Cologne was greatest Major of all time this week. Long live Counter-Strike.Join the discord:https://discord.gg/X3jU4djxUK

NiKo finally wins the Major! Each Falcons' player story line that lead to that moment including karrigan's redemption arch and m0NESY's first Major MVP as well Major impressions and first roster talk in this episode of HLTV Confirmed.➡️ Follow us for updates: https://twitter.com/HLTVconfirmed

On this episode, Joe and Ira talk with Dustin Roddy of Cache River Farms. Dustin focuses on habitat development, both residential and farmland real estate, and has a great deal of experience in these spaces. He covers how to locate and the steps he takes in developing farms to meet certain goals and how he's focused on expediting that process. We enjoyed visiting about hunting styles, habitat and farm development, land purchases, and hunting in both Arkansas and Missouri.    Enjoy!

Become a supporter of this podcast: https://www.spreaker.com/podcast/prepper-broadcasting-network--3295097/support.Support PBN and become a MEMBER of the PBN FAMILY! Free courses, Members only videos, reviews, and podcast! The Prepper's Medical Handbook Build Your Medical Cache – Welcome PBN FamilyJoin the Prepper Broadcasting Network for expert insights on #Survival, #Prepping, #SelfReliance, #OffGridLiving, #Homesteading, #Homestead building, #SelfSufficiency, #Permaculture, #OffGrid solutions, and #SHTF preparedness. With diverse hosts and shows, get practical tips to thrive independently – subscribe now!Newsletter – Welcome PBN FamilyGet Your Free Copy of 50 MUST READ BOOKS TO SURVIVE DOOMSDAYSupport PBN with a Donation 

Welcome to Caching in the NorthWest! This is THE podcast from the birthplace of geocaching, right here in the great Pacific NorthWest.  It's Thursday at 7PM Pacific and we are going to talk about geocaches and geocachers from here and around the globe. So while your GPS is guiding you to a clearing in the middle of a park, we'll be Caching in the NorthWest. We want you to call in your Geocache Log of the Week! Send an email to feedback@CachingNW.com, call into 253-693-TFTC. Call us with your feedback at (253) 693-TFTC Or visit the website at https://CachingNW.com

durée : 00:05:01 - Les Matins de France Culture - par : Alexandra Delbot - Pourquoi un aigle ne chasse-t-il pas comme un faucon ou une chouette ? En analysant les pattes de 37 espèces de rapaces, une nouvelle étude révèle que la forme des phalanges constitue un meilleur indice de leur technique de chasse que leurs célèbres serres. - invités : Myriam Amari Chercheuse doctorante au Muséum national d'Histoire naturelle de Paris, spécialiste des rapaces Vous aimez ce podcast ? Pour écouter tous les épisodes sans limite, rendez-vous sur Radio France

durée : 00:05:14 - Les Matins de France Culture - par : Alexandra Delbot - La perception du champ magnétique est depuis longtemps invoquée pour expliquer le sens de l'orientation des pigeons. Mais où se cache cette boussole ? Selon cette nouvelle étude, des cellules immunitaires chargées en fer dans le foie et la rate les aident à retrouver leur chemin par temps couvert. - invités : Grégoire Loïs Ornithologue, naturaliste au MNHN et directeur adjoint du programme de sciences participatives “Vigie-Nature” Vous aimez ce podcast ? Pour écouter tous les épisodes sans limite, rendez-vous sur Radio France

Gilles Dray, trois enfants, restaurateur et veuf. Il a perdu sa femme dans un tragique accident de voiture. Il avait 31 ans. Et puis la vie a repris son cours, anonyme et tranquille. Sept ans plus tard, la justice est pourtant venue frapper à sa porte. Soudain soupçonné d'assassinat. Retrouvez tous les jours en podcast le décryptage d'un faits divers, d'un crime ou d'une énigme judiciaire par Jean-Alphonse Richard, entouré de spécialistes, et de témoins d'affaires criminelles.Hébergé par Audiomeans. Visitez audiomeans.fr/politique-de-confidentialite pour plus d'informations.

Today we are talking about Drush, Core Contributions, and Drupal's Past with guest Moshe Weitzman. We'll also cover Cache Metrics as our module of the week. For show notes visit: https://www.talkingDrupal.com/556 Topics Moshe Updates and Clients Maintaining Drush Long Term Locale Performance Overhaul CLI in Core Initiative Which Commands Make the Cut Roadmap Contrib Commands Moving Commands Technical Hurdles How to Help From AI Initiative DDEV Add-ons for Local CI MySQL Toolkit Database Images Testing With Real Databases Devel Module Status Organic Groups Origins Where Ideas Come From Finding Drupal Early Days Release Cadence And Backward Compatibility Avoiding Maintainer Burnout Maintaining With AI And Xdebug Resources Drush's Final Act Drupal cli issue DDEV addons https://github.com/ddev/ddev-drupal-contrib https://github.com/weitzman/ddev-mtk https://www.drupal.org/project/dtt Guests Moshe Weitzman - weitzman.github.io moshe-weitzman Hosts Nic Laflin - nLighteneddevelopment.com nicxvan John Picozzi - epam.com johnpicozzi Scott Falconer - managing-ai.com scott-falconer MOTW Correspondent Martin Anderson-Clutz - mandclu.com mandclu Brief description: Have you ever wanted insights into how cache is working on your Drupal site? There's a module for that. Module name/project name: Cache Metrics Brief history How old: created in Oct 2019 by Moshe Weitzman (moshe weitzman), today's guest, a consistent core contributor, a member of the security team, and one of the rare few with a two-digit user id on drupal.org Versions available: 2.0.3, 2.1.0, and 2.2.0, the last of which works with Drupal 8.7.7, 9, 10, and 11 Maintainership Actively maintained Security and test coverage Documentation - in depth README Number of open issues: 2 open issues, 1 of which is a bug, but is marked fixed Usage stats: 37 sites Module features and usage With this module enabled, your Drupal site will log all cache tag invalidations Additionally, cache tag invalidations will be sent to New Relic as custom events, where you can use the rich reporting tools available to mine for further insights. Many Drupal hosting options include New Relic out-of-the-box, and there's a free tier you can use if you're self-hosting, so this a reporting tool lots of Drupal sites can use Cache hits and misses are also sent to New Relic, so you can investigate things like cache misses as a percentage by cache bin Finally, the aforementioned README also includes information about how to use a different analytics provider, in case New Relic doesn't meet your specific needs Drupal sites probably don't need this kind of visibility on a regular basis, but if you're troubleshooting any kind of cache-related issue, this could be really useful

La santé mentale des jeunes se dégrade, et les chiffres sont sans appel : un adolescent sur sept présente un trouble psychique, les diagnostics ont explosé ces dix dernières années, et les hospitalisations ont triplé après le Covid. Pourtant, les moyens restent les mêmes. Pas assez de pédopsychiatres, pas assez de lits, et des parents à qui l'on dit parfois que leur enfant n'est pas “assez malade” pour être pris en charge.Dans cet épisode, je reçois le Dr Cécile Feltin, pédopsychiatre. Avec elle, on décrypte les troubles les plus fréquents chez les jeunes, les facteurs qui influencent leur santé mentale — génétique, attachement, écrans, pression sociale — et le rôle clé des parents, non pas comme thérapeutes, mais comme figures de soutien.On parle aussi des troubles du comportement alimentaire, de l'impact des normes sur le corps.Un épisode pour comprendre, et surtout, pour ne plus rester démuni face à la souffrance des enfants et des adolescents.Au programme :

North Dakota's statewide preparedness system grew out of lessons learned from disasters, public health emergencies, and the realities of serving rural and frontier communities. Tim Wiedrich, director of health response and licensure for the North Dakota Department of Health and Human Services, tells us about how the state developed a unified public-private partnership that supports hospitals, public health agencies, EMS providers, and long-term care facilities across North Dakota.  Tim explains how the events of September 11 and the anthrax attacks reshaped preparedness planning, leading to the creation of a centralized statewide medical cache stocked with critical medical equipment, supplies, pharmaceuticals, generators, and infrastructure support resources. Bridging Systems: Health Care and Public Health Collaborate on Emergency Preparedness in North Dakota | ASTHOPrioritizing Emerging Infectious Disease Cases and Contacts for Follow-Up | ASTHOSubscribe | ASTHO

Welcome to episode 357 of The Cloud Pod, where the weather is always cloudy! Justin and Matt are in the studio this week to bring you all the latest in cloud and AI news! Is AI costing more than the people it replaced? Are CEO's suffering from AI psychosis? Is Opus 4.8 better than 4.7? We answer all of these questions and more this week – so let's get started!  Titles we almost went with this week Valkey Stops Forgetting Your Data Like Your Ex  AI Coding Tools Cost More Than the Coders They Replace Microsoft Discovers AI Budgets Burn Faster Than Enthusiasm Executives Caught Hallucinating About AI Productivity Gains ABBA Said ” Dancing Queen”, but Google Said Data Center AI Now Tells Your AWS Apps How Fragile They Really Are Stop Playing VM Whack-a-Mole With Maintenance Windows Chaos Engineering for Apps Too Scared to Change AWS Rewires the Data Center With One Weird Optical Trick IAM the One Spending All Your Bedrock Money SQL Server Licenses Finally Pack Their Own Bags When AI Hype Meets Productivity Research, It Hurts CEOs Gone Wild: Demos Versus Deployment Reality Serverless Search Finally Learned to Nap Between Requests ElastiCache Finally Remembers Things After a Reboot Valkey Gets Durable So Your Data Stops Ghosting You Zero Data Loss Without Losing Your Microseconds Too Microsoft Build 2026 Scout AI and Quantum Dreams A big thanks to this week's sponsors: There are many cloud cost management tools out there, but only Archera provides insured commitments. It sounds fancy, but it’s really simple. Archera gives you the cost savings of a 1 or 3-year AWS Savings Plan with a commitment as short as 30 days. If you do not use all the cloud resources you have committed to, Archera will literally cover the difference. Other cost management tools may say they offer “insured commitments”, but remember to ask: Will you actually give me my rebate? Because Archera will.  Check out thecloudpod.net/archera to schedule a demo today.  General News  01:45 Microsoft data suggests using AI is more expensive than hiring people: Microsoft canceled most internal Claude Code licenses just months after encouraging widespread adoption, redirecting employees to GitHub Copilot CLI instead.  This does not affect the broader Foundry partnership with Anthropic, but it signals that token costs at scale have become difficult to justify internally. Uber’s situation adds context here: the company reportedly burned through its entire 2026 AI coding tools budget in four months after internal teams were incentivized to compete on usage. This illustrates how adoption incentives can create runaway costs that outpace projected savings.

Pendant l'émission, Sylvie Tellier a toujours un carnet avec elle. Mais que cache-t-il vraiment ? Retrouvez tous les jours le meilleur des Grosses Têtes en podcast sur RTL.fr et l'application RTL.Hébergé par Audiomeans. Visitez audiomeans.fr/politique-de-confidentialite pour plus d'informations.