Clodéric Mars

Well, it is !

@AntoninAtger @clem_corbin Surtout… Le concept de “créateur” d’un système complexe comme Siri est risible en soi.

@clem_corbin On est loin de "cocréateur", voir "créateur de Siri"

La dégringolade de Luc Julia nous apprend beaucoup de choses. ⤵️

@juberti Cool ! I was just wondering of the rt models were "forgotten". However I just tested it on our evaluation dataset (text-to-text) it's roughly 2x slower than the previous model. Is it expected?

New OpenAI Realtime API model drop today! gpt-4o-realtime-preview-2025-06-03 is the latest rev and the vibes are solid, with specific improvements in instruction following and function calling. Check it out on the playground at platform.openai.com/playground/rea….

Would you believe that deep RL can work without replay buffers, target networks, or batch updates? Our recent work gets deep RL agents to learn from a continuous stream of data one sample at a time without storing any sample. Joint work with @Gautham529 and @rupammahmood.

C'est finalement un train à grande vitesse (TGV) qui reliera Québec à Toronto, a appris Radio-Canada. Le fédéral en fera l'annonce dans les prochaines semaines. Jusqu'à 300 km/h. Arrêts à Trois-Rivières, Laval, Montréal, Ottawa et Peterborough ici.radio-canada.ca/nouvelle/21155… #polcan

I vividly remember the time I discovered that ‘std::vector<bool>’ wasn’t actually a vector. I was so pissed off I documented my feelings for posterity cloderic.com/content/2010-0… Anyway, I’m happy to see people are still getting fooled in 2024 :)

@Axelle83211275 Super ! bravo 🎈🎉

Ça y est, officiellement maman de mes enfants !

🪩The @stateofai 2024 has landed! 🪩 Our seventh installment is our biggest and most comprehensive yet, covering everything you *need* to know about research, industry, safety and politics. As ever, here's my director’s cut (+ video tutorial!) 🧵

Being a senior developer is being able to fix the bug in 5 minutes But having the discipline to assign it to a team member knowing it’ll take them 4 hours

the energy i bring to the team

I'm tired of having to prove to people that generative AI systems infringe by default, so here's a megathread of images I've prompted you can use in your discussions. Article by @GaryMarcus and myself at the end. 1/🧵

@jmac_ai @proceduralia I guess my point is that it’s a continuum between planning in known deterministic environments and model free RL. And it’s always the same concepts and objective -> optimizing for a value function.

@jmac_ai @proceduralia I mean we’re saying the same thing, you can “theoretically” build a search tree in any environment, it’s often impractical or impossible, though. RL enables the same thing by learning a model (explicitely for model based implicitely otherwise) from interactions.

Since a number of people asked, this is what I mean by deep learning way of understanding reinforcement learning. It's all about one question: how can I differentiate through the total reward my agent is collecting? The total reward (i.e., the value function) is the objective function of RL. When we see an objective function in deep learning, our goal is to differentiate through it to compute a gradient. You can view many RL techniques and fundamental problems as ways to fulfill that unstoppable desire to differentiate through the value function. If the world your agent is operating in is differentiable you can do it the easy way. If everything is deterministic or reparameterizable à la VAE, you have a complete computational graph. Just compute that gradient and push that value function up. What if you don't have access to the world dynamics? You still want to compute that gradient. Learn a world model from data, then you will generate a valid computational graph and it will give you a gradient to push that value function up. Your world model doesn't work as expected or it's computationally expensive? You can predict the value in a direct way with a critic, a value function approximator, and differentiate through it. With a critic, you can just do one forward and a backward pass. Direct gradient to your agent. No compounding errors, no messed up gradients, nothing too expensive. Learning a critic by just predict the rewards you are observing is requiring too much data? You can use temporal difference, it's a nice concept that exploits the structure of decision-making to speed up learning for a critic. Then you will be able to compute your gradient. You can understand a lot about reinforcement learning as just a quest to compute that gradient of the value function. It's not all of it, but I find it always useful to think about it.

L'entretien que nous venons d'avoir @Cyrielle_Chtl et moi au nom des Ecologistes avec le président de la République est extrêmement préoccupant. Voici en transparence ce que j'en retiens. Comme vous le comprendrez, j'en ressors inquiète, mais pas surprise. On continue ⤵️

Il est donc à la recherche d’un mec (ça on a remarqué) qui serait un opposant MAIS qui lui obéirait en tous points… ce n’est plus la recherche d´un premier ministre mais la chasse au Dahu et donc… ça peut durer longtemps….

Ma mère avait super mal au ventre. La régulatrice du 15 l'a engueulée, infantilisée. Elle est allée à l'hôpital. Elle a été laissée dans un coin pendant des heures avant d'avoir une prise de sang qui a révélé la gravité de sa situation. Ma mère a failli mourir d'une pancréatite

@AIandGames @alexjc @MatthewTJack @nuclai No artificial smarteners then :)

@alexjc @MatthewTJack @nuclai I can confirm that, at present, we sadly don't have a good conference mug. I'm just trying to convince everyone else on the team that we need to address this problem.

So awesome!😃 A much-needed event for video game AI in Europe, organised in a thoughtful manner by great people. It's been over a decade since AIGameDev and Nucl.ai and I'm thrilled we have something new for #gameai #gamedev

We’re thrilled to share that @Autodesk has acquired key intellectual property (IP) and the team from #Golaem. Stay tuned for more amazing things ahead! Learn more on our website: golaem.com/content/news/a… #vfx #crowd #layout #animation

.@lufafarms j’ai un vrai débats à table avec la famille : doit on dire l*ou*fa ou l*u*fa ? J’imagine que les anglophones disent le premier mais avez vous un preference ? #vraiquestion