🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Happy to take curation more seriously as someone labelled as "high taste" in the noisy AI ecosystem. Here are the posts I loved recently, or ones that I think are at least heavily slept on.

Something that comes through clearly in the DeepSeek R1 research paper, and I wish was more broadly understood, is that the DeepSeek researchers really, really want to see themselves as valuable members of the global research community.

My newsletter has been kind of lame for the last two months because I've been spent most of my time on this in-depth explainer on reinforcement learning. I'm proud of the result and I hope people enjoy it.

The two most common questions I get asked about AI are “which AI should I use” and “how do I start using AI?” I wrote a short guide attempting to answer both questions. oneusefulthing.org/p/using-ai-rig…

👌🏼👌🏼👌🏼

The flip side to the concerns about violating copyright in training data is that there is a also a vast trove of important work that nobody reads & where it would be very good if AIs were trained on it. (Most scientific articles, many reports, a lot of old literature & records)

Wow. This is the reasoning the judge used to say that Anthropic training is fair use: "But to make anyone pay specifically for the use of a book each time they read it, each time they recall it from memory, each time they later draw upon it when writing new things in new ways

The Curve 2025 is on! We're bringing together people from a broad range of fields & with disparate perspectives to discuss big questions about AI's future (timelines, natsec, growth, applications, & more) October 3-5 | Berkeley, CA Apply by July 4th for early bird pricing

Okay, heck, this is a panel I need to put some notes down on. 🧵

Today, I testified before the Select Committee on the Chinese Communist Party. I made two key points: 1) the U.S. can win the race to build powerful AI and 2) winning the race is a necessary but not sufficient achievement - we have to get safety right.

This is a big deal. I said I would be surprised if the parliamentarian were willing to sign off on conditioning previously appropriated (and already obligated!) funds on compliance with the AI regulation moratorium under the Byrd Rule. Apparently, so would the parliamentarian.

Useful heuristic when creating a prompt - if I were given this context, could I do the task?

Later in this press conference, Caine says: "We had so many PhDs working on the MOP program, doing modeling and simulation that we were quietly and in a secret way the biggest users of supercomputer hours within the United States of America." Incredible stuff.

Things are getting weird

LLM context windows have grown, but can models really use all this content? We find signs of recent, rapid progress in their ability to do so. Read on to learn more!

METR evaluated a series of recent Qwen and DeepSeek models on our software tasks. We found that the best Qwen models from 2024 perform similar to frontier models from 2023, while DeepSeek models from mid-2025 perform close to frontier models from late 2024.

🚨🚨 Excited to share a new paper led by Haiwen Li with the Community Notes team! LLMs will reshape the information ecosystem. Community Notes offers a promising model for keeping human judgment central but it's an open question how to best integrate LLMs. Thread👇

People keep claiming that US states have proposed 1,000+ AI-related bills this year, but that number just doesn't hold up to analysis. A thread:

While I briefly have no employer, let me tell you what's really happening with AI companies training on public data: [1/n]

This has details I hadn't seen reported before about the President's relationship to the state AI legislation moratorium.