I had a lot of fun chatting with @mattturck from MAD podcast this week! We talked trends in AI, RL and why it unlocks Agents, scaling and so much more: https://t.co/otGWjKGicL links to what we talked about and further reading:

I’m extremely excited about this! Accelerating science and unlocking new breakthroughs is what excites me so much about AI; I would love to hear what capabilities and features would be useful to accelerate your research!

Our position has been consistent: AI will deliver enormous benefits and should be developed thoughtfully. We share the administration's goals of maximizing those benefits, managing risks, and advancing America's lead in AI. A statement from our CEO:

Time for Line 37?

Excited for today $GDLC Bitcoin Ethereum @solana $SOL $XRP @Cardano $ADA

GSV You're Absolutely Right!

As a researcher at a frontier lab I’m often surprised by how unaware of current AI progress public discussions are. I wrote a post to summarize studies of recent progress, and what we should expect in the next 1-2 years: https://t.co/B7438Z9lOF

I strongly recommend signing up - Anthropic is an amazing place to work at, I love the culture, the impact is massive and the pace is incredible! I’ll be at the event, happy to answer your questions there 🤗

Someone today accused me of being anti-muslin. Not true at all. I rather like it.

Anthropic's hosting a social in London in early August for quants & quant engineers interested in a career jump! Especially engineers: if you spend all day optimizing trading system performance, we are excited to chat to you. https://t.co/00YSS1jxTp

New Anthropic research: Alignment faking in large language models. In a series of experiments with Redwood Research, we found that Claude often pretends to have different views during training, while actually maintaining its original preferences.

I'm very excited to announce that I'm joining @AnthropicAI this week, after 10 amazing years at @GoogleDeepMind ! Thank you also to all the amazing people I got to meet and work with, and I'm really looking forward to meeting all my new colleagues 💖 ! https://t.co/hcb5YJR2SB

I was especially excited that it managed to find a proof to P6, the hardest problem this year:

THIS IS THE MOST UNDERVALUED COMPANY ON NASDAQ, according to Warren Buffett’s analysis. Buffett’s rule is simple: Buy real companies that earn real money at irrationally cheap prices. That description fits NextNRG Inc ($NXXT) perfectly. 📊 Facts: * Market Cap ≈ $200–250

Our latest work AlphaProof, building on AlphaZero, LLMs and the @leanprover theorem prover, combined with AlphaGeometry 2 managed to solve 4 IMO problems and achieve silver-medalist level! 🚀 More at

In the normal Turing test, an investigator tries to distinguish a human and an AI. I'm more interested in a stricter version: in each interaction the investigator chooses a preferred response; the AI tries to be chosen as many times as possible. https://t.co/K9dfswkIZw

I’m joining Scale and we are starting a new safety lab! Hiring researchers interested in trustworthy evaluations, red teaming and scalable oversight. These areas require hands-on interaction with human data, and Scale is an unparalleled place to do it.

3. Can we significantly decrease the probability of sampling harmful behaviour by systematically filtering pre-training data? A kind of G-rated dataset?

1. Language models can be thought of as simulators: https://t.co/BYFgaKSn4W 2. The types of agents they can simulate and the likelihood of sampling them depends on the distribution of the training data.

David Van Wie, director of the Johns Hopkins Applied Physics Laboratory—brings more than 40 years of defense-technology experience to students as an instructor in our Engineering for Professionals online master's program in mechanical engineering.

I also wrote a brief summary of our new AlphaTensor paper in my blog:

Also check out the neat cover animation, huge props to our design team!

Excited to announce our latest paper, AlphaTensor, out in Nature today! We describe how we can discover new faster matrix multiplication algorithms using AlphaTensor, a new agent based on AlphaZero. https://t.co/2Nz9pws8Iu https://t.co/iC9H5GE9Dm paper:

Excited to share our latest work on AlphaCode! See our blog post and paper for more: https://t.co/VzOvtehFNw I also wrote up my personal brief summary:

The appearance of glass in this woman suggests a fragility we women know to be real. But Jesus can and does pick up the pieces and put us back together again.