I’m excited to announce I’ve started a YouTube channel. I plan to publish videos regularly explaining my views on AI and its present and future impacts. My first video asks: What happens if there’s an AI crash? https://t.co/sEGeoCyHmk This is my first foray into video (beyond

📢📢 I'm looking for a postdoctoral fellow and so are many of my amazing faculty colleagues @PrincetonCITP. The center's mission is to understand and improve the relationship between tech and society. Apply soon for full consideration. Details: https://t.co/AGphhLkU60 The center

💯 absolutely right! "We think studying the coupling between models and scaffolds is an important research direction going forward, especially as more developers release scaffolds that their models might be finetuned to work well with"

Agent scaffolds are as important as models.

CORE-Bench is solved (using Opus 4.5 with Claude Code) TL;DR: Last week, we released results for Opus 4.5 on CORE-Bench, a benchmark that tests agents on scientific reproducibility tasks. Earlier this week, Nicholas Carlini reached out to share that an updated scaffold that uses

This paper really is groundbreaking. It solves a long-standing embarrassment in machine learning: despite all the hype around deep learning, traditional tree-based methods (XGBoost, CatBoost, random forests, etc) have dominated tabular data—the most common data format in

This is a great piece with some mind-boggling statistics. - At Brown and Harvard, more than 20% of undergraduates are registered as disabled - At Amherst: more than 30 percent - At Stanford: nearly 40 percent Soon, many of these schools "may have more students receiving

I have a guest essay in @nytimes today about autonomous vehicle safety. I wrote it because I’m tired of seeing children die. Done right, we can eliminate car crashes as a leading cause of death in the United States @Waymo recently released data covering nearly 100 million

Many of us intuitively feel that the field of mathematics is going to change, so let's unpack the likely outcomes, without resorting to hyperbole or doomerism.

Right now this "agentic reviewer" is like putting a newspaper online as HTML. It imitates existing referee reports and optimizes agreement with scores, so it learns the current tastes, fads, and biases of human reviewers rather than questioning them. From an economic perspective

I think that Hao made a bad but honest mistake and I don't mean to attack her overall character as a journalist. In contrast, I would like to take this opportunity to directly attack the journalistic integrity of More Perfect Union, who are much more influential in the AI water

NYTimes article quotes someone saying they are "terrified" of Waymo in paragraph 6. Waits until paragraph 33 (out of 44 paragraphs) to mention that they are 91 percent safer than human drivers. How outraged would liberals be if a news outlet covered vaccines like this?

When I was in college at UChicago, I dated another student from Appalachia. She once told me she had gotten straight As in high school calculus -- and when she took the AP exam, she got a 1, the worst possible grade. (And she was pretty talented, she later on pretty well at

Don't make me tap the sign

@snewmanpv @sayashk @DKokotajlo @eli_lifland @thlarsen IMO reliability is only one of the limitations; the others are preference elicitation and the adversarial nature of the environment. I've written a bit about it here: https://t.co/r2lqcoeEQ4 (That said I do think the reliability issue alone is pretty challenging — you'd need ~3

An LLM-generated paper is in the top 17% of ICLR submissions in terms of average reviewer score, having received two 8's. The paper has tons of BS jargon and hallucinated references. Fortunately, one reviewer actually looked at the paper and gave it a zero. 1/3

Glad somebody did this (expert interviews on why LLMs are not currently AGI, and why they could be) feat: @random_walker, @DKokotajlo, @ben_j_todd, @daniel_d_kang, @rohinmshah

Common ground between the authors of AI 2027 and AI as Normal Technology! Coauthored article below.

We enjoyed the opportunity for productive discussion with the authors of AI 2027 to find areas of common ground. We are also planning an “adversarial collaboration”.

https://t.co/cfvtVNanUq

Marc Andreessen has become the avatar of societal decay, the representation of what technology looks like with no vision of the good behind it.