Ben Recht

Why they matter

Researcher with public evidence across AI research, Compute & infrastructure.

AI signals
8
past 30d
Sources
6
distinct domains
Discussions
32
past 30d
Latest signal
2d ago
View every signal from Ben Recht →
Blog: https://argmin.substack.com/ Webpage: https://people.eecs.berkeley.edu/~brecht/ The Irrational Decision: How We Gave Computers the Power to Choose for Us. (Princeton) https://press.princeton.edu/books/hardcover/9780691272443/the-irrational-decision

Articles & links

The full commentary from mathematicians asked to review the proof is thoughtful and worth a read: cdn.openai.com/pdf/74c24085...

cdn.openai.com
AI Weekly's analysis
  • An internal OpenAI model produced a proof disproving Erdős's 1946 unit distance conjecture; the paper posted to OpenAI's CDN calls itself a human-verified version of the AI's work.
  • The AI's original construction beat Erdős's ceiling by only ε ~ 10^-38; Princeton's Will Sawin later pushed the exponent to roughly n^1.014.
  • Scott Aaronson says his former student Lijie Chen 'simply gave GPT the problem' and it returned a several-page argument that held up under human review.
Read full analysis →
View on Bluesky · ♥ 3 ↻ 0 ↩ 1 · 5 from the directory shared this · 109d ago

I don't understand how people can report on companies complaining about distillation and IP theft without mentioning these companies' entire tech stack depends on distillation and IP theft. It's an important detail!

nytimes.com
View on Bluesky · ♥ 44 ↻ 2 ↩ 1 · 5 from the directory shared this · 44d ago

This is probably too mathy, but the first few sections might help?

What is the objective of reasoning with reinforcement learning? arxiv.org
AI Weekly's analysis
  • A new arXiv note argues several RL algorithms for LLMs with binary rewards are stochastic gradient ascent on a monotone transform of one probability.
  • Rejection sampling optimizes the logarithm of that probability; the GRPO algorithm optimizes the arcsine of its square root.
  • The two-sentence abstract by Damek Davis and Benjamin Recht was submitted on October 15, 2025.
Read full analysis →
View on Bluesky · ♥ 9 ↻ 1 ↩ 1 · 7d ago

I hear you, but the families have been willing to share these chats with journalists to raise awareness. See this NYT story, for example. www.nytimes.com/2025/08/26/t...

nytimes.com
View on Bluesky · ♥ 3 ↻ 0 ↩ 1 · 7d ago

Did you see this fun one in the FT about how AI proves Hayek wrong? www.ft.com/content/cd3f...

ft.com
View on Bluesky · ♥ 9 ↻ 0 ↩ 3 · 8d ago

In _The Ideas Letter_, Leif Weatherby, Tyler Shoemaker, and I wrote about the creation of reality through pseudoscience, religion, and psychosis. Of course, the piece is mostly about AI.

Reify This theideasletter.org
View on Bluesky · ♥ 23 ↻ 6 ↩ 4 · 4 from the directory shared this · 87d ago

The arms race of deferring agency and the dark acquiescence to robotic bureaucracy.

argmin.net
View on Bluesky · ♥ 31 ↻ 8 ↩ 2 · 3 from the directory shared this · 31d ago

Adding fuel to the fire, here's a post I wrote about open corpus models last July. It's all still true. www.argmin.net/p/an-open-mi...

argmin.net
View on Bluesky · ♥ 28 ↻ 10 ↩ 1 · 2 from the directory shared this · 44d ago

Now that the entire tech industry is on board, here’s a call to build AI models not just with open weights but with an open corpus.

argmin.net
View on Bluesky · ♥ 51 ↻ 11 ↩ 0 · 2 from the directory shared this · 42d ago

Today on argmin, @jessica.bsky.social's readout of the Public Feedback for AI (& beyond) microconference and how it’s hard to determine who exactly the public is.

argmin.net
View on Bluesky · ♥ 7 ↻ 0 ↩ 0 · 7d ago

Recent commentary

Have to appreciate the audacity of Anthropic defending three successive model releases with diminishing quality by claiming it's both intentional and for our own good.

View on Bluesky · ♥ 30 ↻ 1 ↩ 3 · 89d ago

I'm disappointed that the heated fight between tech bros and medtwitter about midjourney ultrasound snake oil hasn't broken containment over here.

View on Bluesky · ♥ 14 ↻ 1 ↩ 4 · 78d ago

Isn't putting METR & Redwood Research in charge of investigating the OpenAI/Hugging Face incident like putting Irregular in charge of cybersecurity testing?

View on Bluesky · ♥ 12 ↻ 0 ↩ 0 · 11d ago

In Ben Recht's orbit

Center = Ben Recht. Left = members they follow (green edges). Right = members who follow them (blue edges). Top = mutual follows (orange edges, slightly larger). Drag any node to reposition; click to open that profile.

Are you Ben Recht? Show it.

Add the Who’s Who of AI badge to your site or bio. It links back to this profile.

Listed in AI Weekly's Who's Who of AI

Markdown: [![Listed in AI Weekly's Who's Who of AI](https://aiweekly.co/modules/custom/aiweekly_whoswho/images/whoswho-badge.svg)](https://aiweekly.co/whos-who/person/beenwrekt-bsky-social)