Suresh Venkatasubramanian

Nice post by @sophie.huiberts.me - I very much endorse the idea of "evaluate on best three papers" as a way to find a new equilibrium sophie.huiberts.me/blog/2026/me...

There's a new meetup in NYC on Reliable AI! Organizers include Karen Ullrich and @markibrahim.bsky.social. The first is November 16, on synthetic data. I'll be giving one of the keynotes. There's also lightning talks! Sign up and tell the NYC reliable ML community about your work.

so well deserved!

Also, I don't know if there is any way to be clearer, but: If you are a theoretical computer scientist or mathematician currently working for OpenAI, you are not one of the good guys.

It's been a quiet rollout, but ChatGPT has been serving ads since Feb 2026. What's going on behind the scenes? Who's getting those ads and what are they for? And what are the policy implications? www.brookings.edu/articles/und...

This is a good precedent and a more open process. Will be interesting to see how it plays out. And man, ETH!!!

An internal model at OpenAI has produced a claimed refutation of ETH! Russell Impagliazzo, Mohan Paturi, and I have been asked to help organize an open and independent examination of this claim. 1/4

Could not agree with this more.

Powerful thread.

It is interesting that courts are willing to just come out and say that we want human reasoning. Academic disciplines seem institutionally loathe to say the same thing

Sign in and verify via ORCID and OpenAlex for the best experience. Lea is in open beta and is built by @eugenevinitsky.bsky.social and me. It's powered by the AT Protocol, so your social network on Bluesky already exists on Lea, and we can leverage it for the features you see above!

this is absolutely nuts.

Amazed by the videos you can make with Opus 5.5. This is a short video explaining the Browne v. Clarke case (the reggeaton 'Fish Market' case). I wanted a short explainer for class that included the beats without reproducing the songs. I'm pleased with the result.

I saw this claim doing the rounds earlier, via tiktok, and I wanted to repost but knew I should do my due diligence and check the source first. I'm glad I did, because the source is comprehensive and informative. And incredibly relevant to Australia.

i am on a plane sitting next to a lady who looks 40ish and has informed me that this is her first time flying she is visibly excited. she asked the flight attendant how much a soda costs and was happy to learn that drinks are free on airplanes. she's taking lots of pictures. i love this for her

Paging @beenwrekt.bsky.social - tldr is that the result seems plausible

Meet Ruby, CDT’s newest Civil Liberties Watchdog, who doesn’t take orders from anyone. She’s a big believer in reproductive privacy and your right to make personal decisions without government interference. Owner: Elizabeth Seeger

Trying to think of a city that has as dramatic a reveal as Pittsburgh when coming through the tunnel. @costasamaras.com

This is a very very deep snarky dive into the world of chocolate, but it's worth the journey

I need to hand out a copy of this report every time a TSA agent rolls their eyes at me for opting out and says "you know we delete the images"

This is not your typical tech policy conversation. Tech Policy Press and The Privacy Center at Georgetown Law launch "The Edge Dialogues" on Oct. 22, with special guest Dr. Alondra Nelson. First up: exploring what a community-built future looks like, and who holds the power to shape it.

Insurers are introducing exclusions for AI-related losses as AI agents create new risks. Who pays if an employee's rogue agent hacks a site? Insurance companies may become the most effective AI safety regulators because they can refuse to insure companies that don't take safety seriously.

A whole article on the challenge of "only your friends" feeds that doesn't mention bsky or even good ol' Google reader seems woefully clueless about what's possible with tech. Social media algorithms: What if you could turn them off? share.google/UWPlIId6kPeV...

2026 social media in a nutshell.

The other thing that blows my mind is that I haven’t come across anyone talking about sim2real for LM agents. In robotics it is helpful to train in simulation where accidents don’t have consequences. Then fine-tune on reality. Why are we not doing this for LM agents? Don’t tell me it’s too expensive

This is a good thread.

I grew up poor in WV. In my head I am sometimes still that kid. In reality, I work at an elite university. Sometimes important people call me to ask questions. I am clear with myself about the position that puts me in, what I can no longer see and rely on others to keep me in check.

“The current study presents findings from, to our knowledge, the first large-scale randomized experiment evaluating the effects of AI tutoring on student learning at a large U.S. public university.” The only problem is the findings are wrong. brendanbartanen.substack.com/p/ai-tutorin...

I am generally skeptical of the ability to coordinate internationally on technical issues and standards relating to AI. But this kind of story gives me hope. They voted on the leap second! (Sorry leap second). The Guardian share.google/XPF4v5Em4Fsy...

We had an #AIES2026 paper about this! ojs.aaai.org/index.php/AI.... And of course, our AI Policy Summer School is an attempt to communicate the required skills to students :)

More than 30 countries have called for stronger controls on frontier AI and a possible new international institution. Konstantinos Komaitis argues the controls are essential, but AI governance needs an Internet-style network of distributed expertise, not a global overseer.

A constant refrain here at #AIES2026 is "we need to get policymakers to move on the ideas we are presenting". I think that's a deep misunderstanding of the research to policy movement pipeline, and overestimates the "policy readiness" and the structural contributions of individual research papers.

Today, Brown University recognizes Indigenous Peoples Day in honor of Native American and Indigenous peoples and their diverse histories, cultures and contributions.

At the opening session for #AIES2026. And all I can think of is how the opening presentation was clearly Claude generated from the fonts and style :). Nothing wrong with that - it's a time saver - but it's a distinct aesthetic.

In our course and book _Calling Bullshit_, we posit that If something seems too good, or too bad, to be true, it probably is. Now as a college professor I'm pretty confident that overreliance can AI can harm our thinking ability, but just ten minutes of use? Let's dig into this this study a bit.

Independent evaluators are good --- if they are actually independent. www.cnbc.com/2026/10/11/a...

A call to action for senior TCS researchers by @ccanonne.github.io

I'm frankly shocked and pleasantly surprised that the NYT was able to put together a cogent and reasonable argument around AI regulation and liability.

I really enjoyed the clarity of his remarks in this article. Very refreshing.

openai, anthropic, and google are using their most powerful tools to attack public infrastructure

Posting my preliminary proof of P vs NP scribbled on a soiled napkin, attached via cocktail spear to a receipt with the current date and time. Just in case I get scooped by OpenAI.

Anthropic is out with new nonsense about Claude (or the collection of models called Claude) deserving moral consideration and/or be able to feel distress, so my inbox is filling up with journos asking me to comment. I've also got a day job, so here is a brief set of public comments on this: >>