Ce que je verrais dans le programme : Ça serait un mélange de DL, du RL, de de modeling 3D, des LLMs , du signal processing, de l’électronique , de l’embarqué, du réseau.
LLMS
-

OpenAI GPT5-Codex-Max Year End Recap Podcast
By
–
Good time to share that we're releasing some great end of year recap pods every day for your holiday listening! just posted: a great end of year convo recapping @OpenAI Codex and GPT5-Codex-Max with @bfioca and @realchillben
! -
Builders using Anthropic Claude are the future, not emotional support
By
–
Meanwhile all the smartest people I follow are posting about things they are building with Anthropic Claude. The builders are the future. Those who hang onto models for emotional support are not. I'd rather help the builders figure things out. And I know that makes me tone
-
Nostalgia for Sonnet 3.5: AI coding progress evolution
By
–
Do you remember how people losing their minds over Sonnet 3.5 for coding? What innocent times
-
Recommendation of Claude Opus and Codex models for tasks
By
–
Je comprend , et ca va mieux ? Essaye Claude Opus 4.5 ou Codex 5.2 si tu as le budget.
-
Framing consequences for high-stakes LLM responses
By
–
Think about it: When you write an email to a friend → casual
When you write an email that could get you fired → every word matters LLMs learned from both types of text. By framing consequences, you're telling the model: "use the high-stakes mode." -

Why LLMs Respond to Stakes in Text
By
–
You would ask me…Why does this work bro? LLMs are trained on human text and human text is full of stakes. When you add consequences, you're not just giving instructions. You're activating the model's training on how humans think and write when something actually matters.
-
Completing lists on X to improve Grok’s accuracy next year
By
–
My goal with the lists are completeness, which is an impossible task. However, the more complete they are—and they are the most complete of anybody here on X—the better Grok will be able to fill in the gaps next year.
-
MiniMax-M2.1 Excels on SWE-bench and Long-horizon Tasks
By
–
Excels on SWE-bench (incl. multilingual), Terminal-bench, and long-horizon tasks. Try it on Poe across web, desktop, mobile — and via the Poe API at https://
poe.com/MiniMax-M2.1 (2/2) -

MiniMax M2.1: Agent-Optimized AI Model for Coding and Planning
By
–
Now on Poe: MiniMax M2.1 An agent-optimized model built for coding, tool use, and long-horizon planning served via @novita_labs (1/2)