The AI Lectures of @50cent are out!

To choose freedom for oneself and for others is the noblest act of love. Every good parent knows this truth. Thank you @MariaCorinaYA for risking your life to remind us all of this.

We just dropped what we believe is the world's largest study of AI conversations + it found what you talk to AI about has a lot to do with what time it is. @MicrosoftAI researchers found 3 different trends by day, time, and month and 1 rock solid constant

That @demishassabis decided to remain in the UK rather than move to the US was the single most important event in the UK's future. An entire ecosystem is crystallizing around him.

I love this crossbody bag. It is large enough and small enough at the same time. It holds my full size wallet and sunglasses along with anything else that I might need. It isn’t bulky or heavy. It also fits in the Siena Rivera Tote.

Reflection on @dwarkesh_sp's reflections on his interview with Rich Sutton, and why LLMs are exciting because they are Skinnerian, Popperian and Gregorian creatures. Minds with finite capacity cannot adapt forever without having to forget previous knowledge. This is true of

One of the bravest and smartest leaders in the world, @MariaCorinaYA. May Venezuelans be united with their loved ones soon 💛💙❤️ Freedom is a choice, an act of love.

Good researchers obsess over evals The story of Olmo 3 (post-training), told through evals NeurIPS Talk tomorrow. Upper Level Room 2, 10:35AM.

Some really interesting comments below, but the question is still open and requires investigation. I hope a few students pick it up. I liked the discussions on context length generalisation, the fact that we typically train these models as bandits (even when we do RL, which is

Why is it that with ChatGPT, Gemini, Claude, Copilot and other LLMs we have to always start new chats for them to work well? What is the scientific explanation? What are the hypotheses? What is the evidence for each?

Kinda amazing how we can model this. The tectonic movement of the landmasses over 1,000 million years until today (by @nytimes)

Tencent HY 2.0 is officially released. We are rolling out a major performance upgrade to our foundation model, now available via Tencent Cloud API. Built on a Mixture-of-Experts (MoE) architecture (406B total, 32B active parameters) and featuring a 256K context window, HY 2.0

New paper from Qwen team! They showed that because token-level updates are just fragile approximations of sequence rewards, you must use Routing Replay and Clipping to minimize the gap between training & inference for stable RL training in LLMs now trending on AlphaXiv 📈

NeurIPS 2025 Best Paper Award: Attention lets language models decide which tokens matter at each position, but it has limitations—for example, a tendency to over-focus on early tokens regardless of their relevance. Gating mechanisms, which selectively suppress or amplify

I’m looking forward to reading this paper this weekend! @jaseweston and @j_foerst are as creative as it gets.

🤝 New Position Paper !!👤🔄🤖 @j_foerst and I wrote a position piece on what we think is the path to safer superintelligence: co-improvement. Everyone is focused on self-improving AI, but (1) we don't know how to do it yet, and (2) it might be misaligned with humans.

If you’re into triage and AI to improve healthcare, please check out my brother (⁦@jdefreit⁩) and nephew’s startup.

This interesting week started with DeepSeek V3.2! I just wrote up a technical tour of the predecessors and components that led up to this: 🔗 https://t.co/JSAd9cx2s6 - Multi-Head Latent Attention - RLVR - Sparse Attention - Self-Verification - GRPO Updates

Amazing to see AI tools empowering more people to solve complex tasks 👏

With the Excel World Championship underway, I decided to take the M365 Copilot digital challenge. I’m no World Champ… but thanks to Agent Mode, I held my own!

DeepSeek just announced DeepSeek-V3.2 2 open weights LLMs that go head-to-head with SoTA like GPT-5 high, Claude 4.5 Sonnet and Gemini 3.0 Pro With the highlight being their new DeepSeek Sparse Attention SoTA intelligence is now available for everyone to download