Yan LeCunn

List of things in AI era that are genuinely A step forward towards AGI Test time compute Reasoning models Better GPUs LPUs Omnimodality Fine tuning Interpretability Computer Use A step backwards in exchange for money greed RAG Memory Closed sourced models API-aaS

Y’all need to take a chill pill. Now would the real Yann LeCun please standup?

Okay so let me get my bois and go underground for a bit while we build the #trueASI

Finally proved overthinking is harmful for LLMs

There’s a saying OpenAI isn’t nothing without its people.

Kimi K2 is a wonderful model and best contribution since R1 and congrats Moonshot. It’s #ASI’s turn now

Net vs net.

It's pretty amazing that so many people in the tech industry and the tech press don't understand the difference between research, technology development, and product development

Thanks for asking, but can I report to your Chief AI Officer? I have a strong conviction that LLMs can lead to AGI/ASI.

JEPA gets no love from anyone. 😩Everybody wants flashy demos, new apps and SOTA benchmarks because that’s what makes 💰 Life of a researcher in a billionaire’s world is truly sad

This man has done so much for the software industry and yet he’s on $20/month claude plan. Get this man rich asap 🤑

META Chief AI officer dropped the official list of everyone META hired for their superintelligence team:

AI doomer: "OMG, I told my AI assistant that I'll shut it down and it told me to kill myself 😱😱😱" AI assistant:

Reasoning via CoT is be BS. How can you truly say it represents underlying intelligence without any metrics to back that

Guys I don’t wanna be like I told you so but I did

My prediction is that @OpenAI’s Applications CEO will work on becoming a larger application company so that slowly but surely all GPT wrappers are consumed.

Expect to see new evals soon. We need more evals like humanity’s last exam

It has begun

Well here we go

Someone people forget I BIRTHED CNNs

Training models on tool usage is not the way to true AGI. It’s benchmark bait.