Podcasts about BNF

Play Episode Listen Later Aug 13, 2026 20:32

Dès les premiers jours qui suivent la découverte des crimes, la presse s'empare de l'enquête et l'affaire devient un feuillent national. Tandis que certains préjugés brouillent le travail des enquêteurs, le commissaire Sébeille se démène pour faire avancer les investigations. La famille Dominici commence sérieusement à éveiller ses soupçons : le comportement de Gaston Dominici est particulièrement étrange, mais c'est sur Gustave, son fils, que Sébeille concentre son enquête. Ce quatrième épisode propose d'écouter les journalistes dépêchés sur place, notamment le Parisien Jacques Chapus, alors surnommé "le Sherlock Holmes de Lurs", les avocats, l'historien Jean-Yves Royer ou encore Christian Charrier, le producteur du téléfilm L'Affaire Dominici avec Michel Serrault.Crédits archives Les Grandes conférences : éléments bruts, RTF, 05/05/1958. Actualités Méditerranée, ORTF (Marseille), 17/03/1971. L'Oreille en coin du dimanche matin, France Inter, 23/04/1972.Le Vif du sujet, France Culture, 14/10/2003. Le Vif du sujet, France Culture, 28/10/2003.Crédits podcast Un podcast écrit par Florence Dartois - Raconté par Xavier Lemarchand.Restauration et Mixage : Stéphane Rives - Prise de son : Guillaume Solignat, Laurent Thomas, Vincent Dupuis - Unité de production : Céline Amadori, Ameline Tarnagda. Création graphique : Ugo Bouveron - Photo : AFP - Musique(s) : Universal Production Music France - Archives : INA.Cheffe de projet : Lӕtitia Fourmond - Chargée d'édition audio : Edith Monnier - Responsable éditoriale : Zoé Macheret.En partenariat avec RetroNews, le site de presse de la BnF.À retrouver chaque jeudi, à midi, à partir du 30 juillet, sur toutes les plateformes d'écoute et sur podcasts.ina.fr.

Les cosmographes, vers l'infini et au-delà 1/4 : Ronde comme la Terre

France Culture physique

Play Episode Listen Later Aug 10, 2026 56:54

durée : 00:56:54 - LSD, la série documentaire - par : Céline Loozen - La rondeur de la Terre sonne comme une évidence. Mais comment cette connaissance s'est-elle établie et confirmée dans l'histoire, de l'idée à l'observation directe ? - réalisation : Marie Plaçais - invités : Alain Riazuelo Astrophysicien, chercheur CNRS à l'Institut d'astrophysique de Paris (IAP); Catherine Hofmann Conservatrice en chef du département des Cartes et Plans de la BNF; Yaël Nazé Astrophysicienne, maître de recherche au Fonds de la recherche scientifique (FNRS)

plans comme vers lsd terre fonds la terre cnrs ronde cartes bnf marie pla

[Affaire Dominici] EP03 - Les Premières suspicions

Latent Space: The AI Engineer Podcast â€” CodeGen, Agents, Computer Vision, Data Science, AI UX and all things Software 3.0

Play Episode Listen Later Aug 6, 2026 18:20

En fin d'après-midi, le 5 août 1952, le commissaire Sébeille confie une mission prioritaire à ses hommes : retrouver l'arme du crime. Son attention se porte également sur la Grand'Terre, la ferme située à une centaine de mètres de l'endroit où les corps ont été découverts. La propriété appartient à la famille Dominici, des paysans bien connus et respectés dans la région.Les inspecteurs venus de Marseille entreprennent alors de fouiller minutieusement les environs. Ils inspectent la ferme, mais aussi les abords de la Durance, qui coule à proximité. C'est là que sera finalement retrouvée la carabine recherchée. Le soir-même, le commissaire Sébeille fait la connaissance de Gaston Dominici, le patriarche de la famille.Dans ce troisième épisode, journalistes et avocats reviennent sur ce moment charnière de l'affaire. Le commissaire Sébeille raconte ses premiers échanges avec les Dominici et les soupçons qui commencent peu à peu à naître dans son esprit.Crédits archivesLes Grandes conférences : éléments bruts, RTF, 05/05/1958.Aujourd'hui Madame, ORTF, 30/05/1972.Radioscopie, France Inter, 12/03/1973.Le Vif du sujet, France Culture, 14/10/2003.Crédits podcastUn podcast écrit par Florence Dartois - Raconté par Xavier Lemarchand.Restauration et Mixage : Stéphane Rives - Prise de son : Guillaume Solignat, Laurent Thomas, Vincent Dupuis - Unité de production : Céline Amadori, Ameline Tarnagda. Création graphique : Ugo Bouveron - Photo : AFP - Musique(s) : Universal Production Music France - Archives : INA.Cheffe de projet : Lӕtitia Fourmond - Chargée d'édition audio : Edith Monnier - Responsable éditoriale : Zoé Macheret.En partenariat avec RetroNews, le site de presse de la BnF.À retrouver chaque jeudi, à midi, à partir du 30 juillet, sur toutes les plateformes d'écoute et sur podcasts.ina.fr.

crime dans true crime assassins aujourd archives ils premi marseille suspect madame provence suspicion suspects affaire victime restauration france inter cheffe meurtre france culture assassinat bnf rtf cassation jean giono cadavre dominici durance 1952 ortf amadori radioscopie

The Inference Engineering Masterclass — Philip Kiely & Ali Taha, Baseten

Play Episode Listen Later Aug 3, 2026 101:29

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

god amazon ai english google hollywood future training building race video writing fun local mit single open iphone harry potter attention run silicon valley tool focusing speed networking principles pc speech scale couple engineering burning roi masterclass mma pleasure wikipedia chip reflect limits grandma albert einstein expert models intel depending images similar eagle openai mercury fireworks compare inter pipeline nvidia decoding speculation halfway hardware cosmos users api preserving nic generally mp intern smaller layers continuous gpt open source mythos waterloo llama pd apis spectre gram rubin stephen hawking layer clip macbook amd llm 10x medusa sparta fps hopper pip blackwell gpu expense agi cpu kimi tp cache midjourney baseline dsa franken prs dynamo gpus kv spec adjacent continual orpheus tps oof spacetime 2x asics wan speculative t4 4x broadcom itali taha 100x veo gans rl json ethernet rsi asic kling tpm inference ep1 ep2 selo feynman modulation cuda 20b what it takes mlp kiely like you tensor speculators nvidia gpus bnf 25b ampere retrofitting pytorch aie nics sparse ptl yaml mtp itl 8x cerebras jamba minimax glm dsv hbm ai engineer loras tom l zdr sft rubins x6 justifiably tmas martin casado andrew lo gml ptq dgx failure modes nvl nvlink noam brown t4s tp2 tp1

La série INTIMES : entre femmes

L'AFFRANCHIE PODCAST

Play Episode Listen Later Jul 31, 2026 49:08

La série INTIMES : quatre épisodes pour se raconter, en quotidien, en ordinaire, en joies et en orages. Pour nourrir les archives queers, en toute intimité.EPISODE 4 : entre femmes Rencontre avec Alexandre Antolin et Marine Rouch à l'occasion de la parution de leur livre, Entre femmes - Le lesbianisme dans les écrits intimes (1945-1971), aux éditions du Mauconduit.Au sortir de la Deuxième guerre mondiale, l'heure est à la reconstruction nationale : les femmes sont renvoyées au foyer, perçues comme des mères et des épouses contributrices des Trente Glorieuses (1945-1973). Quid de celles qui ne se retrouvent pas dans les normes hétérosexuelles ? les discours scientifiques et moraux condamnent le lesbianisme et continuent d'en faire une perversion. Avant les mouvements féministes, gais et lesbiens des années 1970, lettres et journaux personnels sont le lieu privilégié où les individus peuvent exprimer leurs questionnements et vivre leurs amours. À partir d'archives intimes de ces années d'après-guerre, ce livre explore le « continuum lesbien », selon l'expression d'Adrienne Rich. On y découvrira la correspondance inédite entre l'autrice d'Histoire d'O, Dominique Aury, et son amie et amante Edith Thomas, le journal intime de la jeune Colette Avrane et de Luce Haccard-Perrin ou encore les lettres adressées à Simone de Beauvoir et à Christiane Rochefort â€- Les deux responsables d'ouvrage éclairent chaque ensemble par une mise en contexte historique et partagent leurs propres journaux, dans un geste éthique de réciprocité qui fait résonner ces archives avec les expériences lesbiennes et queer contemporaines. Entre femmes est le 10ème titre de la collection « Vivre/Écrire ». Les textes rassemblés dans ce livre proviennent de l'Association pour l'autobiographie et le Patrimoine Autobiographique (APA), la BNF, l'Imec, les Archives nationales et de collections personnelles.Saison #6 : Univers graphique : Mirion Malle | Habillage sonore : Pierre-Antoine Naline, accompagné de la chorale Dònas d'Òlt d'après le chant La Rota composé par Nadèta Carita | À la conversation et à la réalisation : Soazic Courbet. Hébergé par Acast. Visitez acast.com/privacy pour plus d'informations.

acast archives avant saison vivre visitez femmes rencontre deuxi nad beauvoir quid univers in times bnf trente glorieuses

[Affaire Dominici] EP01 - Une nuit fatale en Provence

Play Episode Listen Later Jul 30, 2026 22:19

Le 5 août 1952, dans le petit village de Lurs, en Provence, une famille anglaise en vacances, les Drummond, est retrouvée assassinée au bord de la nationale 96. Les premiers éléments de la scène de crime recueillis par les gendarmes révèlent des détails troublants : les traces d'un campement rudimentaire, l'identité de Mr Drummond, un scientifique britannique renommé, l'absence de mobile clair et la proximité de la ferme des Dominici, une famille modeste du coin.Ce triple meurtre choque la France et devient rapidement l'Affaire Dominici. Dans ce premier épisode, Maître Émile Pollack, avocat de Gaston Dominici, évoque les zones d'ombre de l'enquête, tandis que l'écrivain Jean Giono décrit le cadre rural et austère de la région, marqué par une histoire violente.Crédits archivesEntretiens avec, France Culture, 15/10/1965.L'Oreille en coin du dimanche matin, France Inter, 23/04/1972.Les dossiers de l'écran, A2 / France 2, 09/09/1980.Crédits podcastUn podcast écrit par Florence Dartois - Raconté par Xavier Lemarchand.Restauration et Mixage : Stéphane Rives - Prise de son : Guillaume Solignat, Laurent Thomas, Vincent Dupuis - Unité de production : Céline Amadori, Ameline Tarnagda. Création graphique : Ugo Bouveron - Photo : AFP - Musique(s) : Universal Production Music France - Archives : INA.Cheffe de projet : Lӕtitia Fourmond - Chargée d'édition audio : Edith Monnier - Responsable éditoriale : Zoé Macheret.En partenariat avec RetroNews, le site de presse de la BnF.À retrouver chaque jeudi, à midi, à partir du 30 juillet, sur toutes les plateformes d'écoute et sur podcasts.ina.fr.

[Affaire Dominici] EP02 - Une scène de crime en péril

Play Episode Listen Later Jul 30, 2026 25:39

Dans la matinée du 5 août 1952, les gendarmes recueillent les premiers indices et témoignages. Mais un obstacle se présente rapidement : l'afflux de badauds, de curieux et de journalistes menace de compromettre la scène de crime que les enquêteurs s'efforcent de protéger.La découverte de douilles près du corps de madame Drummond permet toutefois d'identifier l'arme du crime : une carabine automatique américaine. Un indice essentiel pour la suite de l'enquête.Arrivé de Marseille pour prendre la direction des investigations, le commissaire Sébeille fixe immédiatement sa ligne de conduite : retrouver l'arme du crime, convaincu qu'elle permettra de remonter jusqu'au coupable. Dans ce deuxième épisode, il revient sur ses premières impressions et sur les nombreuses difficultés auxquelles les enquêteurs sont déjà confrontés.Crédits archivesLes Grandes conférences : éléments bruts, RTF, 05/05/1958.Aujourd'hui Madame, ORTF, 30/05/1972.Radioscopie, France Inter, 31/05/1972.Inter actualités de 19H00, France Inter, 22/11/1972.Les dossiers de l'écran, A2 / France 2, 09/09/1980.Le Vif du sujet, France Culture, 14/10/2003.Crédits podcastUn podcast écrit par Florence Dartois - Raconté par Xavier Lemarchand.Restauration et Mixage : Stéphane Rives - Prise de son : Guillaume Solignat, Laurent Thomas, Vincent Dupuis - Unité de production : Céline Amadori, Ameline Tarnagda. Création graphique : Ugo Bouveron - Photo : AFP - Musique(s) : Universal Production Music France - Archives : INA.Cheffe de projet : Lӕtitia Fourmond - Chargée d'édition audio : Edith Monnier - Responsable éditoriale : Zoé Macheret.En partenariat avec RetroNews, le site de presse de la BnF.À retrouver chaque jeudi, à midi, à partir du 30 juillet, sur toutes les plateformes d'écoute et sur podcasts.ina.fr.

crime dans true crime assassins aujourd inter archives marseille suspect madame provence drummond arriv affaire victime restauration france inter cheffe condamn meurtre france culture assassinat bnf franais rtf cassation jean giono cadavre dominici 1952 ortf amadori radioscopie

How to eat sugar without gaining weight (crush your cravings and end energy crashes) | Prof Tim Spector & Prof Sarah Berry

ZOE Science & Nutrition

Play Episode Listen Later Jul 23, 2026 61:07

How can you enjoy sugar without gaining weight, fuelling cravings or crashing later in the day? The answer, surprisingly, is not to cut it out. In today's episode, Professor Tim Spector and Professor Sarah Berry discuss which forms of sugar affect your blood sugar levels most, and what small changes you can make today to reduce the impact. They also explore how sharp blood sugar spikes and dips may affect hunger, energy, inflammation and long-term health. By the end of the episode, you'll know how to spot hidden sugars, how to avoid the afternoon slump and energy crashes, and, most importantly, how to enjoy cake with less impact on your blood sugar. Sugar is not the enemy. But could a few small changes help you enjoy sweet foods without the cycle of cravings, hunger and energy crashes?

[Affaire Dominici] - Bande-annonce

Play Episode Listen Later Jul 9, 2026 2:04

Saison #8 : sortie le 30 juillet 2026.Avec cette nouvelle saison des Archives du crime, l'INA vous propose de plonger dans une affaire qui reste à ce jour l'une des plus plus mystérieuses et sensationnelles affaires criminelles du XXe siècle. Tout commence au début du mois d'août 1952. La France est en vacances. La Provence est écrasée par une chaleur caniculaire. Dans le petit village de Lurs, perché sur les hauteurs des Alpes-de-Haute-Provence, la vie suit son cours, essentiellement rurale, lente, paisible en apparence, entre la récolte de la lavande ou des olives, quelques vignes et la conduite des chèvres au pâturage. Mais dans la nuit du 4 au 5 août, ce calme prend fin avec une tuerie atroce.Au petit matin, trois corps sont retrouvés au bord de la nationale 96 reliant Manosque à Marseille, sur la commune de Lurs.En cet été torride, en manque d'actualités, le fait-divers régional se transforme rapidement en affaire nationale, puis internationale. A la une des journaux, le « Crime de Lurs » devient bientôt « L'affaire Dominici ». Du nom de la famille qui vit à quelques mètres du lieu du drame.Passionnante, l'affaire Dominici l'est sans conteste : le triple meurtre brutal d'une famille étrangère, dont une enfant violemment assassinée, une sommité scientifique, une région de France austère, encore marquée par la guerre, l'absence de mobile évident, et des suspects tout trouvés, une enquête interminable et confuse, nourrie de témoignages contradictoires, de mensonges et d'erreurs d'investigation, accumule les zones d'ombre et instille le doute.Et le mystère s'épaissit encore avec la figure du suspect numéro 1 : Gaston Dominici. Un vieux paysan, chef de son clan, à la personnalité complexe, accueillant et impliqué dans la vie du village, mais aussi taiseux, parfois colérique. Un personnage digne d'un roman de Giono, bientôt emporté, presque broyé, par la machine judiciaire.« L'Affaire Dominici » - 11 épisodes x 30 min. Un podcast écrit par Florence Dartois - Raconté par Xavier Lemarchand.Restauration et Mixage : Stéphane Rives - Prise de son : Guillaume Solignat, Laurent Thomas, Vincent Dupuis - Unité de production : Céline Amadori, Ameline Tarnagda. Création graphique : Ugo Bouveron - Musique(s) : Universal Production Music France - Archives : INACheffe de projet : Lӕtitia Fourmond - Chargée d'édition audio : Edith Monnier - Responsable éditoriale : Zoé Macheret.En partenariat avec RetroNews, le site de presse de la BnF.À retrouver chaque jeudi, à midi, à partir du 30 juillet, sur toutes les plateformes d'écoute et sur podcasts.ina.fr.

Mozart : la folle découverte d'une partition inédite

Lâ€™Heure du Monde

Play Episode Listen Later Jul 7, 2026 25:08

C'est un trésor d'une valeur exceptionnelle : sept pièces pour flûte et harpe, composées en 1778 par Wolfgang Amadeus Mozart (1756-1791), ont été découvertes dans une réserve de la Bibliothèque nationale de France. Une découverte d'autant plus remarquable que le compositeur autrichien affectionnait peu ces deux instruments.C'est en février que François-Pierre Goy, conservateur au sein de l'institution depuis 1995, tombe par hasard sur ces partitions lors d'une vérification de routine. Le musicologue est alors intrigué par plusieurs éléments qui lui rappellent Mozart. L'expertise lève ses doutes : ce cahier est authentifié comme étant de la main du maître, qui l'aurait rédigé à l'âge de 21 ans, alors qu'il est installé à Paris et enseigne la musique à la fille du duc de Guînes.Depuis la découverte, le monde entier s'intéresse à ce trésor retrouvé et les pièces ont été jouées et diffusées, le 22 juin, en première mondiale sur France Musique.Quel est l'itinéraire de ces partitions, vieilles de plus de deux siècles ? Pourquoi Mozart donnait-il des leçons et où en était-il de sa carrière au moment de l'écriture de ces pièces ? Peut-on espérer d'autres découvertes similaires dans le futur ?Dans cet épisode du podcast « L'Heure du Monde », le conservateur François-Pierre Goy raconte avec émotion sa trouvaille et Marie-Aude Roux, journaliste spécialiste de la musique classique au service Culture du Monde, analyse la portée musicale et historique de ces partitions, avant de retracer le parcours de Mozart.Un épisode d'Esther Michon. Réalisation et musiques : Amandine Robillard. Présentation et rédaction en chef : Claire Leys. Dans cet épisode : entretien avec François-Pierre Goy, conservateur à la BNF ; entretien avec Thierry Bodin, expert en autographes musicaux ; archive d'un reportage de TF1 du 26 novembre 2024. Extraits musicaux : interprétation du manuscrit inédit dont les droits sont réservés au département de la Musique de la Bibliothèque nationale de France, l'intégralité de la musique est disponible dans le replay de l'émission « Relax » sur France Musique du 22 juin 2026 ; Die Zauberflöte (« Der Hölle Rache kocht in meinem Herzen ») de W.A. Mozart ; Lacrimosa (Requiem) de W.A. Mozart ; Violin Sonata in G Major, K.301 II.Allegro de W.A. Mozart, Piano Concerto No 23 in A Major, K488 II. Adagio et III.Allegro de W.A. Mozart ; Sonate pour piano No 11 en la majeur, K. 331 (K⁶ 300i), III. Rondo alla Turca de W.A. Mozart ; L'Orfeo, acte 1 de C. Monteverdi ; Konzert in C-Dur für Flöte, Viola da gamba und Cembalo de J.S. Bach.Cet épisode a été publié le 7 juillet 2026. Hébergé par Acast. Visitez acast.com/privacy pour plus d'informations.

Autour de l'exposition Carnets d'Henri Gaudin, itinéraire sensible d'un architecte - À la croisée des disciplines : la pensée de l'architecte Henri Gaudin

acast pens visitez rencontre henri disciplines autour itin architecte sorbonne universit carnets bnf la crois gaudin

Play Episode Listen Later Jul 7, 2026 93:31

Rencontre autour de l'exposition Carnets d'Henri Gaudin, itinéraire sensible d'un architecteConçue en écho à l'exposition Carnets d'Henri Gaudin, itinéraire sensible d'un architecte, cette rencontre réunit artistes, philosophes, théoriciens et chercheurs autour des dimensions structurantes de la pensée et de la pratique d'Henri Gaudin. Architecte, enseignant et théoricien, Henri Gaudin a développé tout au long de sa carrière une pensée humaniste de l'espace, attentive au corps, à l'hospitalité et aux formes architecturales dites « mineures ». Son œuvre bâtie comme ses écrits théoriques ont durablement marqué la réflexion architecturale.Avec Jean-Pierre Le Dantec, théoricien de l'architecture, historien et écrivain ; Catherine Zask, graphiste, typographe, affichiste, artiste, auteure ;Florence Rinaldo, doctorante à Sorbonne Université, chercheuse associée à la BnFTable ronde modérée par Pauline Chougnet, chargée des collections de dessins à la BnF.Séance enregistrée le 11 juin 2026 à la BnF I François-Mitterrand Hébergé par Acast. Visitez acast.com/privacy pour plus d'informations.

L'actualité littéraire de la semaine du 3 juillet 2026

Commedansunlivre

Play Episode Listen Later Jul 3, 2026 0:52

Dans l'actualité littéraire cette semaine : la bibliothèque censurée de Dua Lipa, Albert Camus à la BnF, The Love Hypothesis sur Amazon Prime et la fermeture d'une dizaine de librairies en France.

france amazon prime dans dua lipa litt juillet albert camus la semaine bnf love hypothesis

Critique expos : "Espace entre nous" et "Cartes imaginaires. Inventer des Mondes"

Culture en direct

Play Episode Listen Later Jun 25, 2026 27:45

durée : 00:27:45 - Les émissions culturelles de France Culture - Au menu de ce débat critique, deux expositions parisiennes : un aperçu de la célèbre collection britannique du Wilson Center for Photography à voir au BAL, et une invitation à se perdre avec "Cartes imaginaires, inventer des mondes" à la BNF. - invités : Philippe Azoury Journaliste, critique et auteur, Stéphane Corréard Editorialiste au Journal des Arts Vous aimez ce podcast ? Pour écouter tous les épisodes sans limite, rendez-vous sur Radio France

journal photography critique bal expos radio france corr espace mondes cartes inventer wilson center bnf imaginaires

La Matinale avec nos chroniqueurs Roselyne Bachelot et Hippolyte Pérès : le temps des retrouvailles

Musique matin

Play Episode Listen Later Jun 25, 2026 117:56

durée : 01:57:56 - par : Jean-Baptiste Urbain - D'un côté, un roman choisi par Hippolyte Pérès qui évoque la danseuse Marta Becket et sa rencontre avec un théâtre en pleine Vallée de la mort. De l'autre, la découverte à la BnF d'un manuscrit de Mozart, écrit lors d'un séjour parisien du compositeur dans un lieu bien connu de Roselyne Bachelot. - équipe : Phane Montet, Max Dozolme, Julia Macarez, Valentin Lévy-Chaudet, Morgane Tourreilles, Maxime Laporte Vous aimez ce podcast ? Pour écouter tous les épisodes sans limite, rendez-vous sur Radio France

nos mozart temps le temps radio france vall chroniqueurs retrouvailles bnf matinale hippolyte roselyne bachelot la matinale roselyne

La BD dans tous ses états - Nicolas de Crécy

Wissensnachrichten - Deutschlandfunk Nova

Play Episode Listen Later Jun 24, 2026 73:41

Rencontre avec Nicolas de CrécyUn cycle en salle Ovale, au cœur des collections consacrées au 9e art, invite des auteurs de bande dessinée à dialoguer, sur un mode intimiste, avec lesbibliothécaires et le public. À l'occasion de la sortie de son recueil de dessins Carnets du Japon, réédition augmentée des Carnets de Kyoto, et de son nouveau roman Le Syndrome de Kyoto, la BnF reçoit Nicolas de Crécy afin de discuter avec lui de ses séjours dans l'archipel nippon et de la place particulière qu'occupe ce pays dans son processus de création artistique.Séance enregistrée le 10 juin 2026 à la BnF I Richelieu Hébergé par Acast. Visitez acast.com/privacy pour plus d'informations.

acast tats japon visitez rencontre decr kyoto o vale la bd carnets bnf

La découverte d'un manuscrit inédit de Mozart, coulisses d'une trouvaille musicale

La Revue de Presse

Play Episode Listen Later Jun 23, 2026 7:50

Au sommaire :Dix ans après le départ du Royaume-Uni de l'Union Européenne, quelles sont les conséquences de cette décision ? Malgré les regrets exprimés par une majorité de Britanniques, les dommages économiques sont bien réels. Le Premier ministre Keir Starmer a annoncé sa démission. Il avait amorcé un rapprochement avec l'Europe, mais son mandat a été marqué par de nombreux scandales. Alors que le Royaume-Uni fait son retour dans le programme Erasmus+, l'avenir politique du pays reste incertain.L'enquête administrative sur le traitement de l'affaire Lyhanna confirme qu'il y a eu des dysfonctionnements.Marc Bloch entre au Panthéon. Aux États-Unis, les grands travaux de Donald Trump continuent de susciter la contestation. Sa grande salle de bal coûtera finalement trois fois plus que prévu. La Bibliothèque nationale de France a découvert un manuscrit inédit de MozartHébergé par Audiomeans. Visitez audiomeans.fr/politique-de-confidentialite pour plus d'informations.

La Matinale célèbre la découverte d'un manuscrit inédit de Mozart à la BnF : les coulisses d'une opération hors-norme

Musique matin

Play Episode Listen Later Jun 22, 2026 121:40

durée : 02:01:40 - par : Jean-Baptiste Urbain - Un manuscrit inédit de Mozart vient d'être exhumé à la Bibliothèque nationale de France : une pièce pour flûte et harpe, tout juste enregistrée par deux musiciens de l'Orchestre Philharmonique de Radio France et qui est diffusée pour la première fois dans la Matinale de France Musique aujourd'hui. - équipe : Yassine Bouzar, Max Dozolme, Julia Macarez, Valentin Lévy-Chaudet, Morgane Tourreilles, Maxime Laporte, Clara Michel - invités : Gilles Pécout Président de la Bibliothèque nationale de France, historien, professeur à l'École normale supérieure de la rue d'Ulm, Sibyle Veil Présidente-directrice générale de Radio France, Michel Orier Directeur de la Musique à Radio France Vous aimez ce podcast ? Pour écouter tous les épisodes sans limite, rendez-vous sur Radio France

pr france mozart ration hors radio france norme les coulisses biblioth coulisses bnf matinale france musique la matinale gilles p yassine bouzar

Haie und Mantas, Japan verrutscht, Philosophie der Emotionen, Mozartstücke entde

Play Episode Listen Later Jun 19, 2026 5:02

+++ Haie nehmen Mantas zum Kratzen +++ Japan rutscht sechs Millimeter nach Osten +++ Politik braucht Emotionen +++ Unbekannte Mozartstücke entdeckt +++**********Weiterführende Quellen zu dieser Folge:Chafing behavior by Galapagos sharks on oceanic manta rays in North America's largest marine protected area, Marine Biodiversity, 01.04.2026ScS-triggered slip on megathrust interfaces after the 2011 MW 9.0 Tohoku-Oki earthquake, Science, 18.06.2026Die Philosophie der Gefühle: Wie Emotionen unser Weltverhältnis prägen, Interview auf der Website der Uni, Zugriff: 19.06.2026Découverte d'un manuscrit autographe inédit de Mozart au département de la Musique de la BnF, Pressemitteilung der Französischen Nationalbibliothek, 19.06.2026**********Ihr könnt uns auch auf diesen Kanälen folgen: TikTok und Instagram .

tiktok science interview japan north america gef ihr emotionen mozart franz philosophie musique quellen galapagos mw zugriff haie millimeter pressemitteilung bnf mantas nationalbibliothek

Lyhanna : l'autopsie confirme la thèse du viol - Le 5/5 d'Amandine Bégot

C à vous

Play Episode Listen Later Jun 19, 2026 13:44

Le 5/5 d'Amandine Bégot : 1 jeune sur 3 a déjà conduit sous l'emprise de la drogue ou de l'alcool 12 % des jeunes ont déjà consommé du protoxyde d'azote Association Antoine Alleno : du rap au service de la prévention PSG : Achraf Hakimi bientôt jugé pour viol Lyhanna : l'autopsie confirme la thèse du viol l'ADN de jérôme Barella retrouvée sur le corps de Lyhanna Lyhanna : son corps aurait été bâillonné avec du scotch Lyhanna : des traces d'ecchymoses retrouvées sur le corps Emmanuel Macron évoque “un continent caché” de violences sexuelles Ukraine : 3 morts dans des frappes russes Russie : Poutine plus isolé que jamais Ukraine : Loukachenko présente ses excuses à Zelensky Un nouveau manuscrit de Mozart découvert à la BNF Fête de la musique : la Philharmonie jouera des inédits de MozartCoupe du monde : Le monde entier met l'ambiance aux Etats-Unis Les belles histoires de la Coupe du mondeNew York met à l'honneur ses basketteurs Coupe du monde : la star de la première semaine Backroom : quand les spectateurs font du doublageTous les vendredi et samedi vers 19h40 sur France 5, Amandine Bégot vous fait un point sur l'actualité du jour dans son “5 sur 5”.

new york tiktok france ukraine mozart emmanuel macron zelensky exact viol etats unis coupe backroom la th philharmonie bnf confirme france tv barella amandine b

Biblioteca Nacional de França mostra mapas portugueses como objectos de poder, imaginação e controlo

Artes

Play Episode Listen Later Jun 17, 2026 9:51

A Biblioteca Nacional de França organiza uma exposição sobre mapas, mostrando alguns exemplares portugueses do século XVI. São mapas que mostram o Mundo conhecido, mas que representam também o desconhecido, com todos os mitos e já algumas marcas de crenças que se vão perpetuar nos séculos seguintes marcados pelo colonialismo. Um mapa serve para orientar, guiar e percorrer um caminho até ao ponto de chegada. Olhamos para os mapas, agora nos nossos telemóveis, para nos assegurarmos que fazemos o percurso mais curto ou traçamos a rota mais directa. Mas tempos houve em que os mapas misturavam mundos fantásticos, certezas matemáticas e vontades políticas. Na exposição “Cartes Imaginaires” ou “Mapas Imaginários”, patente na Biblioteca Nacional de França, ou BnF, em Paris, até 19 de Julho, o departamento de mapas desta instituição retraça os primórdios dos nossos mapas modernos, com exemplares desde o século XIV até aos mapas reinterpretados por artistas contemporâneos. Na primeira parte da exposição, chamada “Mundos Inexplorados” figuram grandes cartógrafos portugueses ou ao serviço dos reis portugueses. Entre o acervo mostrado ao público está o Atlas Miller, um atlas que data de 1519 e foi elaborado pelos cartógrafos Lopo Homem, Pedro Reinel e Jorge Reinel ilustrado pelo miniaturista António de Holanda. Este mapa exaustivo para a sua época traça o Oceano Atlântico Norte, a Europa do Norte, Madagáscar, ou ainda o Brasil. Terá sido oferecido pelo Rei D.Manuel a Francisco I de França. Para as comissárias da exposição Julie Garel-Grislin et Cristina Ion, este é um trabalho de precisão, confirmado com certezas matemáticas e também imaginação à mistura. “Eles faziam bem o seu trabalho. O que é interessante sobre esta exposição, e falando especificamente sobre os mapas portugueses e, em particular, temos o Atlas de Miller, que foi produzido em 1519, é que podemos ver que há mapas que são muito precisos e outros que são muito menos. É verdade que a precisão dos mapas que apresentamos é muito desigual No entanto, por exemplo, no que diz respeito às linhas costeiras, exploradas já há muitos anos pelos ocidentais, temos uma linha bastante precisa e ela também é o resultado do trabalho colaborativo de todos os navegadores”, explicou a comsisária Julie Garel-Grislin. “Esta cartografia marítima era, acima de tudo, uma cartografia de experiências, portanto não era, estritamente falando, uma cartografia matemática, como a cartografia que seria herdada da chegada ao Ocidente da geografia de Ptolomeu. Mas isso não significa que as duas tradições sejam mutuamente exclusivas; pelo contrário, a tradição da cartografia matemática e a tradição da cartografia, digamos, empírica, realmente complementam-se e enriqueceram-se mutuamente. Elas interagiam e os atlas que circularam com a geografia de Ptolomeu tinham ido buscar referências da cartografia marítima. Já os mapas náuticos foram buscar à geografia matemática questões de escala ou o clima”, complementou Cristina Ion. “Se pensarmos no continente africano, é preciso saber que os mapas feitos pelos portugueses, em particular, foram retomados, reformulados e assim por diante. E isto aconteceu até ao período moderno. Os mapas feitos pelos portugueses tinham uma enorme influência, o que significa que teríamos um mapa que seria copiado e disseminado entre todos os cartógrafos europeus. Portanto, no que diz respeito ao Ocidente, os mapas portugueses não são influentes, é algo diferente, Eles são como uma matriz para os outros cartógrafos”, reforçou Julie Garel-Grislin. Mais do que indicar caminhos, o Atlas Miller mostra a fauna e a flora dos continentes assim como as projecções mais ou menos correctas sobre os habitantes de terras longínquas. Estas interpretações do outro levaram a preconceitos que se cristalizaram durante séculos sobre a estratificação dos seres humanos e a capacidades dos povos europeus a reconhecerem a humanidade dos povos autóctones sobretudo em África e na América. “Estas representações não foram produzidas inicialmente para justificar ou legitimar a escravidão, pois elas são muito mais antigas do que esse flagelo. Obviamente, essas representações têm outras origens. Elas derivam de textos antigos, mas é verdade que elas ajudaram a construir uma certa maneira de olhar para a África, para as pessoas que vivem lá, e, claro, num certo sentido, a tornar a escravidão talvez mais fácil de pensar, de aceitar, de praticar para os países colonizadores”, afirmou Julie Garel-Grislin. “A presença dessas criaturas híbridas nos mapas deve-se principalmente a textos antigos, particularmente à “História Natural” de Plínio, o Velho, escrita no I século depois de Cristo e que foi transmitida pelo enciclopedismo medieval e que levou a colocar monstros nos mapas. Isso não era, de forma alguma, uma maneira de desumanizar os seres humanos, mas, ao contrário, de mostrar como os monstros faziam parte da criação. Eles não estavam fora do Mundo, estavam no Mundo. Depois, durante o período conhecido como as Grandes Descobertas, começou a haver um encontro real e efectivo entre os europeus e os povos indígenas, especialmente do continente americano. Com toda a ambiguidade e o debate que surgia já nos textos da época, à volta da natureza, desses povos que eram ao mesmo tempo adornados com todas as virtudes da antiguidade e considerados como selvagens que comiam carne humana e ignoravam a salvação da alma. Mesmo nessa altura já havia debates que denunciam a violência da colonização europeia e, paradoxalmente, em última análise, a escravidão em larga escala e a exploração colonial maciça de outros continentes pelos europeus”, explicou Cristina Ion. Se no início do período das grandes descobertas, estas representações tinham um sentido quase religioso e a necessidade de representar o desconhecido, rapidamente, e com a prática da escravatura em massa, muitos destes mapas foram sendo alterados e substituídos por grandes espaços brancos. Espaços brancos convidavam à aventura, à conquista, ou seja, à colonização. Nos séculos XV e XVI, os mapas constituíam um símbolo de conhecimento e estatuto de um país, mas sobretudo uma ferramenta de poder para os monarcas em pleno período de Descobertas. “É um facto. Sabemos muito bem que os mapas, especialmente naquela época, eram um instrumento essencial de poder. Claro que os vemos como obras de prestígio, mas eles são instrumentos de Estado, instrumentos de poder, e também podemos ver isso, por exemplo, na questão do Tratado de Tordesilhas. Agora sabemos que, aparentemente, os portugueses estavam mais bem informados do que aparentavam estar”, detalhou Julie Garel-Grislin. “Os mapas eram estritamente controlados pelos Estados e pelos reinos ibéricos, que estavam envolvidos numa competição aguerrida pelo controlo das rotas marítimas. E os mapas eram segredos de Estado. De facto, eles eram guardados a sete chaves e protegidos de espiões, sendo copiados, transmitidos e transportados em segredo”, indicou Cristina Ion. Para a exposição “Mapas Imaginários”, foi preciso escolher entre cerca de um milhão de mapas, 10 mil atlas e 300 globos. Um trabalho de supressão permanente com explica a comissária Julie Garel-Grislin. “Organizar uma exposição foi uma experiência de privação e supressão. Portanto, foi extremamente difícil escolher. Foi mesmo muito complicado. A ideia era encontrar mapas que fossem suficientemente inteligíveis em si mesmos, que também tivessem um apelo estético interessante e que os mapas pudessem interagir uns com os outros. Ou seja, que se encaixassem no itinerário da exposição, que se encaixassem no objectivo intelectual que definimos com a Cristina e que, colocados lado a lado, constituíssem esse diálogo”, declarou Julie Garel-Grislin. A Biblioteca Nacional de França faz um esforço quotidiano de preservação destas obras, muitas delas com mais de cinco séculos, o que significa um investimento permanente nas condições de acondicionamento, no restauro e na conservação. Recentemente, o Departamento de Mapas e Planos mudou-se para as novas instalações, na renovada Biblioteca Richelieu. Esta obra custou quase 270 milhões de euros e as obras duraram 12 anos, com a mudança das colecções para este palácio a ser chamada “a mudança do século” já que foi preciso transportar 42 milhões de objectos entre mapas, moedas ou livros raros e preciosos. “Tem sido um grande esforço nos últimos 10 anos na configuração de um lugar de armazenamento. O departamento de mapas e planos foi transferido para a recém-restaurada Biblioteca Richelieu e aí, nessas novas instalações temos 14 quilómetros para armazenar e bem acondicionar os nossos mapas”, concluiu Cristina Ion. A exposição “Mapas Imaginários”, que inclui também a história de como muitos autores de ficção e não só criaram os Mundos para os seus livros, está patente na Biblioteca François Mitterrand, em Paris, até dia 19 de Julho.

Une histoire de la BnF - De Richelieu à Tolbiac : déplacements, controverses, émotions

Play Episode Listen Later Jun 8, 2026 54:51

Demi-journée d'étude coordonnée par Emmanuelle BermèsPorté conjointement par la Bibliothèque nationale de France et le Comité d'histoire du ministère de la Culture, le séminaire « Une histoire de la BnF » propose de poser un regard extérieur sur l'histoire de l'établissement, en interrogeant ses liens avec les autres acteurs culturels et l'histoire des politiques publiques.Du projet architectural à l'informatisation du catalogue en passant par la répartition des collections, l'ouverture du site de Tolbiac ne se fit pas sans heurts ni controverses. Trente ans après, cette séance proposera d'esquisser une histoire émotionnelle de la « Très Grande Bibliothèque ».Séance coordonnée par Emmanuelle Bermès (École nationale des chartes), enregistrée le 29 mai 2026 à la BnF I François-Mitterrand Hébergé par Acast. Visitez acast.com/privacy pour plus d'informations.

culture france acast visitez comit motions une histoire placements biblioth trente richelieu bnf controverses tolbiac grande biblioth

Construction d'une salle de tribunal dans une prison, Roumanie touchée par des drones russes et rénovation du site François Mitterrand de la BNF

Le journal de 8h00

Play Episode Listen Later May 29, 2026 8:25

Au sommaire :Annonce de la construction d'une salle de tribunal dans la prison de Fleury-Mérogis pour juger les détenus les plus dangereux, dans le cadre de la lutte contre la criminalité organisée.Survol du territoire roumain, membre de l'OTAN, par un drone russe qui s'est écrasé sur un immeuble, soulignant les tensions géopolitiques en Europe de l'Est.Bénéfices records de TotalEnergies au premier trimestre, relançant le débat sur la taxation des superprofits des entreprises pétrolières.Rapport sénatorial alarmant sur l'urgence de rénover le site François Mitterrand de la Bibliothèque Nationale de France, vieillissant et coûteux.Impact économique potentiellement important des canicules à venir, avec des pertes de PIB estimées jusqu'à 7% pour les pays les plus exposés.Hébergé par Audiomeans. Visitez audiomeans.fr/politique-de-confidentialite pour plus d'informations.

Les visiteurs du soir - De la Forêt des Rêves bleus au Pays d'Oz : pérégrinations cartographiques en littérature jeunesse

Play Episode Listen Later May 20, 2026 56:18

Rencontre organisée par le Centre national de la littérature pour la jeunesse de la BnFDans le cadre du cycle « Les visiteurs du soir », le Centre national de la littérature pour la jeunesse (CNLJ) de la BnF propose des rencontres régulières avec des professionnels du livre et de l'enfance. La littérature pour la jeunesse entretient avec la carte un rapport singulier. Son rôle n'est pas d'orienter le lecteur dans le dédale des mondes fictionnels, mais de l'inviter à en dépasser les frontières. Cette conférence propose de plonger dans certaines de ces cartes, du panorama du monde enchanté de Bernard Sleigh aux cartes de Narnia ou de Bilbo le Hobbit.Séance enregistrée le 5 mai 2026 à la BnF I François-Mitterrand. Hébergé par Acast. Visitez acast.com/privacy pour plus d'informations.

acast pays hobbit visitez rencontre narnia litt jeunesse bilbo bleus mitterrand bnf les visiteurs

Dominique Perrault

L'heure bleue

Play Episode Listen Later May 12, 2026 49:24

durée : 00:49:24 - La 20e heure - par : Eva Bester - L'architecte et urbaniste spécialiste du "groundscape" Dominique Perrault, auréolé du prestigieux prix Mies van der Rohe pour la BnF en 1997, voit paraitre une imposante Monographie de plus de 440 pages qui lui est consacrée dans la Collection beaux Livres de chez Gallimard. - réalisation : Lola Costantini, Céline Villegas, Fanny Leroy, Franck Olivar, Juliette Lorphelin, Anaïs Boucher - invités : Dominique Perrault Architecte de la Bibliothèque François Mitterrand Vous aimez ce podcast ? Pour écouter tous les épisodes sans limite, rendez-vous sur Radio France

collection livres radio france boucher villegas mies biblioth gallimard perrault rohe bnf monographie dominique perrault

Bodybuilders, masques et artifices avec Valérie Belin

Culture en direct

Play Episode Listen Later Apr 25, 2026 108:43

durée : 01:48:43 - Comme un samedi - par : Arnaud Laporte - Une carte blanche autour de l'image, de la photographie, du visage, avec l'artiste plasticienne Valérie Belin et ses invité·e·s. - réalisation : Alexandre Fougeron - invités : Valérie Belin Artiste plasticienne et photographe; Camille Morineau Historienne de l'art, directrice de l'association Aware (Archives of Women Artists, Research and Exhibitions); Héloïse Conésa Conservatrice en charge de la photographie contemporaine au département des Estampes et de la photographie de la BnF; Jean-Pierre Criqui Historien d'art; Aure Chanteuse Folk

research comme bodybuilders exhibitions masques belin avec val bnf women artists estampes artifices

Les bibliothèques européennes au secours des LLM

DeepTechs

Play Episode Listen Later Apr 5, 2026 45:21

Et si la qualité des données valait mieux que leur quantité ? C'est le pari audacieux que défend Rémi Daudin, directeur scientifique de Forvia, l'équipementier automobile né de la fusion de Faurecia et HELLA. Physicien de formation, passé par la physique des particules, l'industrie pétrolière et l'automobile, il porte un projet qui bouscule les fondements mêmes du développement des grands modèles de langage (LLM) : les entraîner sur les fonds des bibliothèques nationales plutôt que sur le web ouvert.« Les données du web, c'est de la soupe », résume-t-il sans détour. Publicités, opinions non sourcées, contenus dupliqués — les LLM actuels ingèrent en masse des informations de mauvaise qualité, ce qui les contraint à gonfler leurs paramètres pour absorber cette complexité. Résultat : des modèles lourds, coûteux, et potentiellement biaisés. « Si 10 % du corpus affirme que la Terre est plate, le modèle le répètera une fois sur dix », illustre-t-il. Sa solution ? Se tourner vers les bibliothèques. Un livre, contrairement à une page web, est un raisonnement construit, vérifié, légal. La Bibliothèque nationale de France, qui numérise ses collections depuis trente ans, compte dix millions d'œuvres. Selon Rémi Daudin, s'entraîner sur un tel corpus permettrait de produire des modèles 100 fois plus efficaces, plus petits, moins énergivores — et non biaisés.Le projet, baptisé Mirandola — en référence à Jean Pic de la Mirandole, l'humaniste italien qui aspirait à tout savoir — se veut aussi un acte de souveraineté culturelle. À l'heure où les algorithmes américains façonnent l'information mondiale, l'idée est de redonner à l'Europe la maîtrise de ses valeurs et de sa vérité. L'Académie française, la BNF et des décideurs politiques seraient déjà sensibles à la démarche. Encore en phase d'amorçage, Mirandola cherche ses partenaires techniques — possiblement un acteur européen du LLM — pour une première phase estimée entre six et douze mois. Hébergé par Acast. Visitez acast.com/privacy pour plus d'informations.

france acast europ encore visitez terre llm hella ques publicit biblioth secours bnf la biblioth mirandola physicien faurecia forvia

#156: You Can Make a Fortune Trading - Don't Let Anyone Tell You Otherwise - With Matt “Pax” Kenah

The AlphaMind Podcast

Play Episode Listen Later Mar 10, 2026 72:18 Transcription Available

In 2000, a young Japanese trader named Takashi Kotegawa (better known as BNF) began trading small-cap stocks from his bedroom with just $13,500. Eight years later, he had turned that modest stake into a staggering $153 million.This legendary feat serves as the anchor for a recent viral X post by veteran trader Ryan Scott (aka The Flow Horse). Ryan's post (linked below) argues a provocative truth: despite the cynics, it remains entirely possible to build a fortune in the markets. But if the opportunity is real, why do so few actually succeed?In this episode, we welcome back veteran trader Matt "Pax" Kenah to dissect this debate. We explore the deep psychological, mental, and behavioral hurdles all traders face, specifically focusing on the challenges of transitioning to a home-trading environment and the discipline required to become one of the elite few who truly thrive.Links: Ryan Scott (Horse) @TheFlowHorse - The Post: You Can Make a Fortune Trading - Don't Let Anyone Tell You OtherwiseFind out more about Matt and the PAX group at https://thepaxgroup.org/You can follow Matt on Twitter at https://twitter.com/paxtrader777AlphaMind: Powering Performance in Global MarketsAlphaMind brings powerful change, growth and development to people and businesses within global markets. Driven by a deep understanding of how markets work, and how people and businesses function within them, we partner with clients to create personal performance improvements that elevate returns across their trading activities. Go to the AlphaMind website to know more.

japanese investing fortune behavior driven trading stock market pax hedge funds day trading bnf swing trading price action ryan scott trading psychology market profile

Et Casanova inventa Casanova 7/11 : Casanova, figure moderne ou conservatrice ?

figure histoire moderne casanova mauvais france culture xviiie bnf inventa les nuits marie andr giacomo casanova

Play Episode Listen Later Feb 28, 2026 110:47

durée : 01:50:47 - Les Nuits de France Culture - par : Albane Penaranda - En 2010, la BnF acquiert le manuscrit original d'"Histoire de ma vie" de Giacomo Casanova, marquant un événement culturel majeur. À cette occasion, l'émission "Mauvais genres" revient sur la vie du séducteur vénitien, dévoilant son opportunisme, ses multiples conquêtes et son héritage littéraire. - réalisation : Massimo Bellini - invités : Michel Delon Professeur émérite de littérature française du XVIIIe siècle à l'Université Paris IV-Sorbonne; Noëlle Châtelet Ecrivaine, sociologue, comédienne; Patrick Wald Lasowski; Emmanuel Pierrat Ecrivain, essayiste et avocat; Anaïs Kien Productrice déléguée; Jean-Baptiste Thoret Historien et critique de cinéma, réalisateur; Marie-Andrée de Saint-André Chroniqueuse

La légende Barbara s'expose à la Bibliothèque nationale de France

De vive(s) voix

Play Episode Listen Later Feb 11, 2026 29:00

La Biblitohèque nationale de France (BnF) consacre en ce moment, depuis le 27 janvier 2026 et ce jusqu'au 5 avril 2026, une exposition à la célèbre chanteuse et parolière française Barbara, au sein de sa fameuse Galerie des Donateurs, nommée « Dis, quand reviendras-tu ? Barbara et son public ». De son vivant, Barbara n'a jamais songé à archiver sa propre légende, compliquant ainsi la tâche à ceux qui auraient voulu retracer le parcours atypique, tantôt sulfureux, tantôt idyllique, de la compositrice de « L'Aigle noir ». Mais, grâce à ses admirateurs et notamment aux dons de l'association Barbara Perlimpinpin, la BnF a pu se procurer une centaine de documents faisant référence au lien tout particulier que l'autrice-compositrice entretenait avec ses fans. Photos, vidéos, partitions, costumes de scène, tenues de ville, notes, programmes, autographes, courriers... Tant d'items rares que les intéressés peuvent dès maintenant consulter à la BnF, afin d'analyser sa méthodologie de travail et de mieux comprendre les sentiments quasi-amoureux que Barbara éprouvait pour son public. Ma plus belle histoire d'amour, c'est vous. Au fil de l'exposition, les visiteurs découvriront son rapport singulier à la scène, depuis ses débuts dans de modestes cabarets belges jusqu'à ses dernières apparitions publiques en 1993. À la BnF, la légende Barbara perdure, donc, le long d'un cheminement d'exposition intime, et pourrait se résumer en une seule phrase prononcée par la vedette elle-même sur la scène de Bobino en 1966 : « Ma plus belle histoire d'amour, c'est vous ». Et, comme tous les mercredis, Lucie Bouteloup décrypte une expression bien connue de la langue française dans sa chronique « La puce à l'oreille » avec la complicité de la lexicographe Géraldine Moinard des éditions Le Robert et des élèves de CM1B de l'École élémentaire Arago du 14ème arrondissement de Paris. Et, cette semaine, Lucie vous a à l'œil ! Invitée : Coline Arnaud, coordinatrice du développement des publics, département de la Musique, BnF. Programmation musicale : Barbara et ses titres « Dis, quand reviendras-tu ? », « Sid'amour à mort » et « Perlimpinpin ».

culture france photos musique tant invit dis nationale galerie programmation bnf la biblioth bobino arago connaissances culture

La légende Barbara s'expose à la Bibliothèque nationale de France

De vive(s) voix

Play Episode Listen Later Feb 11, 2026 29:00

culture france photos musique tant invit dis nationale galerie programmation bnf la biblioth bobino arago connaissances culture

[REDIFFUSION] 78. J. R. R. Tolkien et l'histoire, avec Emilie Fissier et Frédéric Manfrin

Paroles d'histoire

Play Episode Listen Later Feb 9, 2026 54:51

Les invité-e-s : Emilie Fissier et Frédéric Manfrin, du département d'histoire de la Bibliothèque Nationale de France, commissaire associée et commissaire principal (avec Vincent Ferré) de l'exposition TolkienL'événement : exposition « Tolkien, voyage en terre du Milieu », du 22 octobre 2019 au 16 février 2020 à la Bibliothèque Nationale de France.La discussion :Les origines de l'exposition « Tolkien, voyage en Terre du milieu », en lien avec la Bodleian library d'Oxford, et le « Tolkien estate » (1:15)Les choix d'objets mis en regard des œuvres de Tolkien (5:50)Un parti-pris de l'exposition : ne rien montrer de postérieur à 1972, pour replonger les visiteurs dans l'imaginaire propre à l'auteur (7:20)Le genre de la « fantasy », déjà en partie constitué quand Tolkien commence à écrire (8:20)Les anneaux de Tolkien ne sont pas le Ring de Wagner ! (10:15)L'enfance de Tolkien près de Birmingham, et la sensibilité à la nature, aux paysages, qui en découle (12:40)Le rapport complexe de Tolkien à Shakespeare, et à l'antiquité gréco-latine (15:10)L'invention linguistique comme source fondamentale de son inspiration (17:00)L'entrée en guerre de 1914, moment ambigu pour qui travaille sur les langues et l'aire germanique (19:10)La marque de la Grande Guerre sur l'œuvre de Tolkien, travaillée par la mort (21:00)Tolkien dans l'entre-deux-guerres, savant et écrivain pour ses enfants (24:20)Son talent graphique et la variété de sa palette (26:00)Le succès du Hobbit (1937) et le début d'une véritable carrière d'écrivain (28:30)Le travail propre de Tolkien sur la langue anglaise, et sa musicalité (30:00)La cosmogonie de la Terre du Milieu (31:45)La réception du Seigneur des anneaux, et son ampleur sur les campus américains dans les années 1960 en particulier (32:45)Les paradoxes d'une lecture pacifiste de Tolkien, alors qu'un personnage comme Faramir souligne la légitimité de la guerre (34:25)Le Moyen âge de Tolkien, antérieur à la conquête normande, et loin de la matière arthurienne (37:00)La juxtaposition de périodes et de régions dans le monde imaginaire de Tolkien : Minas Tirith, allusion à Byzance (41:20)Le thème de la quête, fonctionnant de manière inversée dans le Hobbit et le Seigneur des anneaux (42:40)Un Tolkien « médiéviste » qui va jusqu'à inventer une tradition manuscrite de son propre texte ! (44:10)Les sources d'inspiration de Tolkien pour les créatures fantastiques dont il peuple son œuvre (45:15)Un Tolkien qui ne sépare pas les créatures en « races » (48:00)L'apparence des manuscrits de Tolkien, qui évoquent à leur façon le Moyen âge (50:10)Quels objets, quelles œuvres ont le plus marqué les commissaires de l'exposition ? (51:50)Pour aller plus loin :Tolkien, voyage en Terre du Milieu, catalogue de l'exposition de la BNF, 2018.John Garth, Tolkien et la Grande Guerre, Paris, Christian Bourgois, 2014.Un podcast créé, animé et produit par André Loez et distribué par Binge Audio. Contact pub : project@binge.audioHébergé par Audiomeans. Visitez audiomeans.fr/politique-de-confidentialite pour plus d'informations.

france ring shakespeare birmingham wagner tolkien quels hobbit visitez terre nationale milieu jrr tolkien moyen seigneur rediffusion biblioth audiomeans bnf le moyen grande guerre binge audio faramir bodleian byzance christian bourgois

ENTRETIEN - Depuis quand les rues sont éclairées en France ? - Avec Sophie Reculin

Nota Bene

Play Episode Listen Later Feb 6, 2026 99:27

L'éclairage public, de nos jours, il y en a à beaucoup d'endroits, et ce n'est pas rare qu'il soit éteint aux heures creuses de la nuit pour lutter contre la pollution lumineuse. Mais à ses débuts, c'était une toute autre histoire, parce qu'il a été très mal reçu par les habitants ! Et quand je parle de ses débuts, ce n'est pas le 19e siècle avec l'électricité, car l'éclairage public, il est né en France dans la deuxième moitié du 17e siècle ! Mais alors, comment ça s'est fait ? Eh bien pour le découvrir, j'ai eu le plaisir de recevoir dans un nouvel entretien historique l'historienne Sophie Reculin, dont c'est la spécialité, pour qu'elle nous éclaire à ce sujet ! Je vous souhaite une bonne écoute sur Nota Bene !➤ Pour approfondir le sujet, découvrez : ➜ Le livre de Sophie “L'invention de l'éclairage public en France. De la nuit illuminée à la nuit éclairée (1697-1789)” : https://www.septentrion.com/FR/livre/?GCOI=27574100084770➜ Sa page Academia : https://univ-lille.academia.edu/SophieReculin

france night police acast arte academia quand depuis temps histoire sont visitez en france mus entretien xixe vaincre xviiie rues jean no avec sophie xviie bnf concordance early modern europe nota bene sophie l carnavalet empire a history gcoi

Le grand poète Baudelaire, l'exilé

Franck Ferrand raconte...

Play Episode Listen Later Jan 23, 2026 23:46

visitez legrand baudelaire charles baudelaire rimbaud audiomeans bnf

Critique expo : "Les Mondes de Colette" à la BNF François-Mitterrand, entre lumière et zones d'ombre

Culture en direct

Play Episode Listen Later Nov 28, 2025 9:50

durée : 00:09:50 - Les Midis de Culture - par : Marie Labory - "Les Mondes de Colette" à la BnF François-Mitterrand explore la vie et l'œuvre de l'écrivaine à travers manuscrits, photographies et objets, révélant ses multiples facettes et ses univers littéraires. - réalisation : Laurence Malonda - invités : Céline du Chéné Productrice à France Culture; Joseph Ghosn Directeur adjoint de la rédaction de Madame Figaro

culture expo critique zones lumi mondes mitterrand bnf les mondes madame figaro laurence malonda

Napoléon Bonaparte au siège de Toulon, la naissance d'un stratège

Au cœur de l'histoire

Play Episode Listen Later Nov 25, 2025 13:42

En juin 1793, alors que la France s'enfonce dans la Terreur, la ville de Toulon tombe aux mains des royalistes, qui font alliance avec les Britanniques et les Espagnols contre les républicains révolutionnaires. Pour le Comité de Salut public, organe mis en place par la Convention nationale afin de restaurer l'autorité du gouvernement, il est essentiel de reprendre la ville. Entre alors en scène un jeune capitaine d'artillerie corse répondant au nom de Napoléon Bonaparte (1769-1821).Dès le mois de septembre, Bonaparte est à Toulon, engagé au sein des troupes de la Convention pour mener le siège qui doit aboutir à la reprise de la ville. Chef d'artillerie, il conçoit des plans pour faire tomber des forts stratégiques et parvient à galvaniser ses troupes. En décembre, après quatre mois de siège, l'assaut final est donné. C'est le début de l'ascension de celui qui deviendra empereur une décennie plus tard. (rediffusion)Au Cœur de l'Histoire est un podcast Europe 1.- Présentation et écriture : Virginie Girod- Production : Armelle Thiberge et Morgane Vianey- Réalisation : Nicolas Gaspard- Composition du générique : Julien Tharaud- Visuel : Sidonie Mangin Bibliographie :- Thierry Lentz et Fanny Farieux, Napoléon, la biographie inattendue, Passés Composés, 2021- Charles-Eloi Vial, Napoléon, la certitude et l'ambition, Perrin et BnF, 2020.Ressources en ligne :- Le souper de Beaucaire - Gallica, BnF- Essai sur les idées de jeunesse de Bonaparte - Fondation Napoléon- Les blessures de Napoléon - Fondation NapoléonHébergé par Audiomeans. Visitez audiomeans.fr/politique-de-confidentialite pour plus d'informations.

europe pr france chefs pass convention visitez comit strat salut perrin bonaparte napol toulon la naissance compos terreur britanniques audiomeans auc bnf espagnols

À la table du Goncourt

Le goût du monde

Play Episode Listen Later Nov 8, 2025 28:59

Direction Drouant, place Gaillon au centre de Paris. Drouant, institution parisienne, cœur battant de la littérature puisque dans ces murs, à deux pas de l'Opéra, se réunit l'académie Goncourt. 10 membres élus à vie décernent depuis plus d'un siècle le prix éponyme, au meilleur livre ouvrage en français de l'année. Un prix voulu par les frères Goncourt, à la fin du XIXème, et devenu un incontournable de la vie littéraire française. Un fauteuil et un couvert Singularité de l'académie Goncourt, ses membres ont un fauteuil, comme pour les académies de l'Institut de France, mais n'étant pas sous la coupole, mais dans un restaurant et la table ayant son importance, les «Goncourt» ont également un couvert. Ils se retrouvent une fois par mois le mardi et le premier mardi de novembre pour décerner le prix. Ce jour-ci, un menu spécial est pensé, conçu et servi aux académiciens, au lauréat ou à la lauréate et son éditeur. Sa composition est gardée secrète pour être une surprise et une fête le jour du prix. Son élaboration est le fruit du travail complice, curieux, passionné de James Ney, le directeur de Drouant et du chef des cuisines du restaurant, Romain Van Thieunen depuis 2022. La littérature pour inspiration, un «supplément d'âme» apporté, au-delà des plats servis. Ce menu est un cadeau et un trésor de créativité. Notre tournage s'est fait en amont, sous le sceau de ce secret, et notre émission patiemment gardée pour vous la proposer au bon moment, comme vous pouvez aussi si vous le souhaitez aller le savourer chez Drouant. Avec James Ney, directeur de Drouant, et Romain Van Thienen, chef du restaurant. Le Goncourt a été décerné le 4 novembre 2025 à Laurent Mauvignier pour «La maison vide» aux éditions de Minuit. Le prix Renaudot a été décerné le même jour à Adélaïde de Clermont-Tonnerre pour «Je voulais vivre». Dans les coulisses et du côté de la presse, nous avons voulu savoir comment était vécue cette journée et ce que représentait le prix Goncourt. Question posée à l'une des figures de la littérature sur RFI : la journaliste Catherine Fruchon-Toussaint, productrice de l'émission Littérature sans frontières, un mot enfin du prix Goncourt des détenus et du choix Goncourt à l'international. À écouter Littérature sans frontières. Pour découvrir l'histoire du restaurant Drouant, fondé en 1880 par Charles Drouant, 16-18 place Gaillon, Paris 2ème arrondissement. Le menu Goncourt est disponible chez Drouant jusqu'au 15 décembre 2025. Il est conseillé de réserver 72 heures en amont. Pendant l'émission, est diffusé un extrait de la remise du prix Goncourt de 1954, et l'élection de Jean Giono au siège de Colette. Le menu du prix Goncourt 2025 placé sous le signe de Colette : langoustine, huîtres et caviar, lièvre à la royale, langues d'oursin, cerises fossilisées, et en dessert une pomme flétrie opaline de caramel, lamelles de poire crues. «Rien n'est plus émouvant qu'un fruit qui vieillit à la lumière.» Colette. Merci à Emma Morris chez Viennot sans qui cette émission n'aurait pu exister. Pour aller plus loin - Le site de l'académie Goncourt - La maison vide, de Laurent Mauvignier - Bernard Clavel – Les fruits de l'hiver - Huysmans – À rebours - Les mandarins, de Simone de Beauvoir. - Colette : Pour découvrir la femme, l'autrice, à Paris, la BNF organise une grande exposition autour de l'auteure. - Chez Virginie à Montmartre. Programmation musicale : Drama Queen, de Melba. En images

france dans ils pendant rien xix litt gastronomie beauvoir alimentation rfi drama queens minuit programmation goncourt la table singularit bnf jean giono renaudot clermont tonnerre viennot

#138 Lucie Moriceau-Chastagner (musée de l'Armée) Collections photo et images de conflits

Les Voix de la Photo

Play Episode Listen Later Oct 29, 2025 64:42

Dans cet épisode, nous recevons Lucie Moriceau-Chastagner, historienne de l'art et responsable des collections photographiques au Musée de l'Armée. Après un parcours au Service historique de la Défense (SHD), puis à l'Établissement de communication et de production audiovisuelle de la Défense (ECPAD), elle a rejoint en 2021 le musée de l'Armée.Spécialiste des images de guerre – qu'elles soient historiques ou contemporaines –, elle revient sur son expérience de valorisation de ce patrimoine complexe, entre conservation, recherche et exposition. Lucie nous parle également de la résidence photographique du musée de l'Armée, lancée en 2022, qui ouvre un dialogue entre photographes contemporains et mémoire militaire.Bonne écoute !00:28 – Présentation de Lucie Moriceau-Chastagner : historienne de l'art, responsable des collections photo au Musée de l'Armée, ancienne responsable de la médiation à l'ECPAD.01:23 – Définir la photographie militaire : représentations du fait militaire par militaires, civils, amateurs et studios.08:12 – Début du parcours académique : école du Louvre, recherche et muséologie ; stage à la BNF auprès de Maxime Préaud.17:40 – Premiers travaux au Service historique de la Défense (SHD) : conseil aux chercheurs et constitution de dossiers d'œuvres.22:15 – Articulation SHD / ECPAD : complémentarité entre archives et production audiovisuelle.23:05 – Treize ans à l'ECPAD : évolution du statut, concours et missions documentaires.30:38 – Arrivée au Musée de l'Armée en 2021 et missions de conservatrice des collections photo (plus de 100 000 phototypes).34:34 – Lancement de la résidence photographique du musée de l'Armée en 2022, en lien avec l'exposition Photographie en guerre.39:48 – Constitution du jury et premières candidatures, dont Alice Broyer et Guillaume Herbaut.53:49 – Conseils aux jeunes chercheurs : ouverture disciplinaire, histoire de la photographie et approche critique des images de conflits.55:03 – Exposition Photographie en guerre et acquisitions récentes, notamment de Laurent Van der Stockt.01:02:00 – Travail en réseau et circulation des projets au sein des musées et festivals de photographie.01:03:00 – Invitation à découvrir les collections et la résidence photographique au musée de l'Armée.Site du musée de l'Armée : https://www.musee-armee.fr/accueil.htmlet lien vers la résidence photographique : https://www.musee-armee.fr/minerve-le-musee-de-demain/residence-photographique.htmlLien vers la liste d'attente de ma formation Du projet à l'objet : les bons réflexes pour réussir votre livre photo : https://forms.gle/eG9svk51rrYhhJ7X8Lien vers mon questionnaire pour vous aider à faire un point sur votre carrière artistique : https://bit.ly/carriereartistiqueLien vers mon questionnaire pour vous aider à faire un point sur votre projet de livre : https://bit.ly/LVDLPlivrephotoMon site : https://marinelefort.fr/Pour vous inscrire à la newsletter du podcast : https://bit.ly/lesvoixdelaphotonewsletterLe site du podcast : https://lesvoixdelaphoto.fr/Et vous pouvez retrouvez le podcast sur Instagram, Facebook et LinkedIn @lesvoixdelaphoto Hébergé par Acast. Visitez acast.com/privacy pour plus d'informations.

pr service invitation dans acast constitution images bonne visitez travail louvre collections mus premiers conseils arriv lancement photographie conflits bnf treize

Les Pénélopes, une épopée féministe sur internet

La marche du monde

Play Episode Listen Later Oct 18, 2025 48:29

La Marche du monde vous invite cette semaine à découvrir Les Pénélopes, une épopée féministe sur internet, un nouvel épisode documentaire consacré aux Pénélopes, un média féministe créé par des Françaises dans le sillage de la 4è conférence mondiale sur les femmes à Pékin qui se tenait en septembre 1995, organisée sous l'égide de l'organisation des Nations unies (ONU). Pékin, septembre 1995, pendant 10 jours, des dizaines de milliers de femmes venues du monde entier, des déléguées gouvernementales, des représentantes d'ONG, des militantes féministes, étaient rassemblées pour travailler à la rédaction d'une plateforme d'action en douze points pour l'égalité femmes-hommes. En attendant la conférence suivante annoncée en 2000 à New York, des femmes ont décidé d'agir et de profiter de l'émergence d'internet pour travailler en réseau et relayer en ligne des informations produites par et pour des femmes. C'est le cas des Pénélopes, un groupe de Françaises qui créent sur le web des médias féministes à destination du monde entier. On s'est dit : les médias en France ne font pas leur travail. Il faut se mettre au travail, c'est à nous de le faire parce qu'il n'y aura personne d'autre pour le faire à notre place. […] On voulait du transversal, de l'horizontal, de l'international surtout. Donc on va faire notre magazine féministe international sur le web. Joëlle Palmieri, co-fondatrice des Pénélopes Alors qu'internet arrive à peine dans les foyers, et que les premiers sites d'informations voient le jour, les Pénélopes décident de leur emboîter le pas et de tisser leur toile féministe. Le site www.penelopes.org sera lancé en janvier 1997. Au programme : des articles en français, en anglais et en espagnol, sur la vie des femmes et sur les luttes des femmes à l'international, mais aussi la création d'une Web TV et d'une web Radio avec les moyens du bord, de la formation toute azimut auprès de nombreux collectifs de femmes, et enfin une présence à de nombreuses conférences et rendez-vous internationaux, de Séoul, à Dakar, en passant par Porto Alegre pour les premiers forums sociaux. Avec Pénélope, on faisait des choses qu'on n'avait jamais faites avant et rien n'était impossible. […] C'était rare de rencontrer des militantes féministes du monde entier ; il fallait récolter les témoignages et apprendre les unes des autres. Malin Bjork, membre des Pénélopes 80 ans après la création de l'ONU, le 24 octobre 1945 à San Francisco, qui avait accordé une place importante aux droits des femmes, avec l'affirmation du principe d'égalité femmes-hommes dans le préambule de la Charte, et 30 ans après la conférence des femmes à Pékin en 1995, La marche du monde vous emmène à la rencontre des Pénélopes, ces pionnières du web féministe. Aux sons de leurs archives numérisées récemment par Bibliothèque nationale de France (BNF), Les Pénélopes, une épopée féministe sur internet, c'est un nouvel épisode de la Marche du monde signé Joëlle Palmieri et Maxime Grember, réalisé par Sophie Janin et produit par Valérie Nivelon. Avec les témoignages de : - Malin Bjork, membre des Pénélopes- Elsa Boulet, fondatrice des Pénélopes- Marie-Hélène Mottin-Sylla, partenaire des Pénélopes- Joëlle Palmieri, fondatrice des Pénélopes.Remerciements à : - Dominique Foufelle, fondatrice des Pénélopes- Lin McDewitt Pugh, partenaire des Pénélopes- Julie Guillaumot, cheffe du service vidéo à la BNF- Nicole Fernandez Ferrer, déléguée générale du Centre audiovisuel Simone de Beauvoir- Laurence Sarniguet et Eugénie Ducret de la Sonothèque de RFI.Musiques : - Computer Love, de Kraftwerk- Woman, de Neneh Cherry- People has the power, de Patti Smith.Ouvrages : - Les Pénélopes : un féminisme politique 1996-2004, de Dominique Foudelle et Joelle Palmieri, 2014Archives : - Journal France inter, 5 septembre 1995 (archives INA),- Émission Cyberfemmes du 28 avril 1999, «Spéciale Pénélopes»,- Émission Cyberfemmes du 4 juin 2000 à New York, «Pékin+5 : des femmes prennent les médias en main»,- Émission Cyberfemmes du 9 juin 2000 à New York, «La Francophonie en péril ?»,- Reportage du 16 décembre 2000 à Dakar pendant le sommet de la dette, - Reportage du 26 janvier 2002 à Porto Alegre lors du 2e Forum social mondial, «Manifestation contre le sexisme et l'impérialisme»,- Reportage du 26 janvier 2001 à Porto Alegre lors du 1er Forum social mondial, «Manifestation pour l'avortement»,- Reportage du 13 avril 2002 à Paris sur le cinquième anniversaire du magazine en ligne des Pénélopes; concert du groupe Les Zarmazones.,Ressources : - Site internet des Pénélopes- Web radio, “Voix de femmes”- Émissions Cyberfemmes, Catalogue Bnf,- Exposition «Europes en partage» du 13 septembre 2025 au 6 septembre 2026 au musée de la Bnf, présentation d'extraits de l'émission Cyberfemmes du 6 octobre 1999, intitulée «L'Europe et les femmes».Carrousel photos

Trading Legends Series Part 2: Dennis, BNF & Zanger

Talking Wealth Podcast: Stock Market Trading and Investing Education | Wealth Creation | Expert Share Market Analysis

Play Episode Listen Later Oct 16, 2025 41:06

Fil and Pedro return with round two of Trading Legends, revealing how traders Richard Dennis, BNF & Dan Zanger turned small stakes into hundreds of millions. They'll unpack the exact strategies and mindsets that helped these retail traders defy the odds and consistently win. If you're chasing the real secrets behind extraordinary trading success, this episode is one you can't afford to miss.

trading fil bnf zanger legends series richard dennis

Les visionnaires : Sasha Luccioni, la chercheuse qui concilie IA et climat (3/4)

La Loupe

Play Episode Listen Later Oct 15, 2025 13:34

Alors que la planète est traversée par de multiples conflits ces dernières années, que nos sociétés paraissent de plus en plus divisées, et que les innovations comme l'intelligence artificielle peuvent être sources d'inquiétude, L'Express a choisi de sortir de cette vision pessimiste. De la géopolitique à la science, en passant par l'éducation et la culture, nombreux sont les hommes et les femmes à révolutionner le monde de demain. Cette semaine, La Loupe brosse le portrait de quatre de ces visionnaires. Dans ce troisième épisode, on s'intéresse aux recherches de Sasha Luccioni, une scientifique qui tente d'introduire plus de transparence dans l'IA.Retrouvez tous les détails de l'épisode ici et inscrivez-vous à notre newsletter.L'équipe :Écriture et présentation : Charlotte BarisMontage et réalisation : Jules KrotLecture et traductions : Anne-Charlotte PhanCrédits : France Inter, Monde numérique, Trench Tech, BnF, TedX, NBC News.Musique et habillage : Emmanuel Herschon / Studio TorrentLogo : Jérémy CambourPour nous écrire : laloupe@lexpress.fr Hébergé par Acast. Visitez acast.com/privacy pour plus d'informations.

tedx dans acast alors monde nbc news musique visitez climat france inter chercheuse bnf sasha luccioni

Tentation autoritaire chez les jeunes / Nouveau leadership climatique / Les mille et une vies de Colette

Les matins

Play Episode Listen Later Sep 27, 2025 120:38

durée : 02:00:38 - Les Matins du samedi - par : Nicolas Herbeaux - Au programme cette semaine : la jeunesse française serait de plus en plus attirée par un pouvoir politique autoritaire ; 10 ans après l'Accord de Paris, qui incarne le nouveau leadership climatique ? ; en dernière partie, visite guidée dans l'exposition "Les mondes de Colette" à la BNF à Paris. - réalisation : Jean-Christophe Francis - invités : Marie Boëton Journaliste à La Croix; Thibaud Voïta Docteur en science politique, conseiller à l'Institut Jacques Delors et co-directeur de master à l'EM Lyon; Jean-Daniel Collomb Professeur d'études américaines à l'Université Grenoble Alpes; Laurence Le Bras conservatrice en chef, cheffe du service des Manuscrits modernes et contemporains au département des Manuscrits de la BNF

leadership chez nouveau mille jeunes vies docteur les jeunes climatique tentation bnf autoritaire les mille manuscrits les matins jean christophe francis

400. Les cathares, réalités d'une persécution, avec Arnaud Fossier

Paroles d'histoire

Play Episode Listen Later Sep 15, 2025 56:55

L'invité : Arnaud Fossier, MCF à l'université de BourgogneLe livre : Les Cathares, ennemis de l'intérieur, La fabrique, 2025.La discussion :· Pourquoi un livre sur les Cathares sous forme de synthèse critique ? (1:00)· Les étapes historiographiques de la question cathare (9:30)· Les incertitudes documentaires, et l'exemple de la « charte de Niquinta » (18:20)· Le cœur du livre : les « cathares » sont une construction de l'Église, mais des « dissidents » ont bien été persécutés (24:30)· Que peut-on savoir des rites pratiqués par ces « dissidents » ? (36:00)· Les différentes phases de persécution (39:00) Le conseil de lecture : Roberto Saviano, FalconeIllustration: Saint Augustin combattant les hérétiques, Bnf fr 241, 38vUn podcast créé, animé et produit par André Loez et distribué par Binge Audio. Contact pub : project@binge.audioHébergé par Audiomeans. Visitez audiomeans.fr/politique-de-confidentialite pour plus d'informations.

visitez arnaud alit audiomeans bnf roberto saviano binge audio mcf

Du fonds Radio France à Gallica : quand les archives inspirent la fiction

Les matins

Play Episode Listen Later Jul 25, 2025 40:14

durée : 00:40:14 - France Culture va plus loin (l'Invité(e) des Matins d'été) - par : Astrid de Villaines, Stéphanie Villeneuve, Sarah Masson - Réunis pour une visite exceptionnelle de l'entrepôt des archives de Radio France, le romancier Philippe Jaenada, Nola N'Diaye, du projet Gallica de la BNF et Thomas Henry, responsable des collections de la documentation à Radio France, échangent sur les enjeux et les usages de l'archive. - réalisation : Félicie Faugère - invités : Nola N'Diaye Adjointe au chef du service de la coopération numérique de Gallica; Philippe Jaenada Écrivain français; Thomas Henry Responsable des collections de la documentation à Radio France

france radio fiction quand archives fonds radio france villeneuve france culture bnf matins les archives philippe jaenada gallica thomas henry villaines

NOTA BENE - Quel lien entre Napoléon et la Franc-maçonnerie ?

Nota Bene

Play Episode Listen Later Jul 13, 2025 21:23

Mes chers camarades, bien le bonjour !Ce n'est pas rare d'entendre que la Franc-maçonnerie va de paire avec la République, et pourtant, il y a une grande part d'idées reçues là-dedans ! Idées reçues qui ne sont d'ailleurs pas neuves du tout, puisqu'elles se sont forgées peu après la Révolution française. Pourtant, on le sait maintenant, cette idée est largement fausse, et cette association n'est d'ailleurs pas toujours allée de soi. Aujourd'hui, on va le montrer en s'intéressant plus particulièrement aux relations de cette organisation avec un autre régime politique, le Premier Empire, au début du XIXe siècle. Si vous vous attendez à des théories du complot sur des satanistes, des reptiliens, ou des judéo-bolchéviks, vous allez être déçus. Mais si vous voulez en savoir plus sur cette mystérieuse organisation qu'est la Franc-maçonnerie et sur une page importante de son histoire, alors vous êtes au bon endroit !Bonne écoute !

Passion Égypte, une histoire 3/4 : Champollion, déchiffrer l'homme

Le Cours de l'histoire

Play Episode Listen Later Jul 3, 2025 51:53

durée : 00:51:53 - Le Cours de l'histoire - par : Xavier Mauduit, Maïwenn Guiziou - En 1822, Jean-François Champollion, égyptologue prodige de trente-deux ans, perçait le secret des hiéroglyphes. Comment cette découverte révolutionnaire a-t-elle permis de renouveler les connaissances sur l'Égypte ancienne et de fonder l'égyptologie moderne ? - réalisation : Alexandre Manzanares - invités : Vanessa Desclaux Conservatrice à la BNF en charge des collections orientales, des manuscrits Égypte antique et Proche-Orient chrétien; Karine Madrigal Égyptologue, chargée de l'inventaire et de l'étude du fonds Champollion conservé aux archives départementales de l'Isère

passion histoire jean fran une histoire proche orient bnf le cours champollion xavier mauduit alexandre manzanares

Les Éditions de Minuit, esprit d'autonomie 7/14 : "Jérôme Lindon défend le Nouveau Roman avec Robbe-Grillet et dénonce la guerre d'Algérie avec Vidal-Naquet"

Play Episode Listen Later Jun 8, 2025 41:01

durée : 00:41:01 - Les Nuits de France Culture - par : Albane Penaranda - Anne Simonin, auteur d'un ouvrage intitulé "Les Éditions de Minuit, 1942-1955, le devoir d'insoumission", et Anne Renoult de la BnF, retracent les combats de Jérôme Lindon, directeur des Éditions de Minuit de 1948 à 2001. Entretien 2/3 par Albane Penaranda. - réalisation : Virginie Mourthé - invités : Anne Renoult Conservatrice à la Bibliothèque Nationale de France et commissaire d'exposition; Anne Simonin Historienne au CNRS

france nouveau guerre vidal nationale la guerre entretien esprit le nouveau cnrs biblioth fend france culture minuit nonce bnf les nuits lindon robbe

Les Éditions de Minuit, esprit d'autonomie 14/14 : Anne Simonin : "Les Éditions de Minuit sont nées libres et le sont restées"

Play Episode Listen Later Jun 8, 2025 5:02

durée : 00:05:02 - Les Nuits de France Culture - par : Albane Penaranda - Pour guider nos pas dans l'histoire des Éditions de Minuit, nous sommes en compagnie d'Anne Renoult, commissaire de l'exposition "Les Combats de Minuit" à la BnF et Anne Simonin, auteur de l'essai "Les Éditions de Minuit, 1942-1955, le devoir d'insoumission". Entretien 3/3 par Albane Penaranda. - réalisation : Virginie Mourthé - invités : Anne Renoult Conservatrice à la Bibliothèque Nationale de France et commissaire d'exposition; Anne Simonin Historienne au CNRS

france sont nationale libres entretien esprit cnrs biblioth france culture minuit bnf les nuits simonin

Les Éditions de Minuit, esprit d'autonomie 1/14 : La naissance des Éditions de Minuit en 1941 : un acte de résistance