Boaz Barak
@boazbaraktcs
Followers
27K
Following
12K
Media
808
Statuses
9K
Computer Scientist. See also https://t.co/EXWR5k634w . @harvard @openai opinions my own.
Cambridge, MA
Joined January 2020
Paper allows to verify and so reduce apparent entropy of inference. They use it to block model weight exfiltration, but it's a natural question that I guess will find other applications too. First author is Harvard's own @RoyRinberg !
New paper! A compromised inference server can leak model weights by hiding them in normal-looking responses๐ฅท But LLM inference is nearly deterministic, so we can verify outputs with a trusted server & detect this (and inference bugs) This slows data exfiltration by >200x ๐งต
1
3
30
OK, I think GPT 5.2 Pro is actually a step change in usefulness for my applications (algebraic geometry/number theory research).
33
54
729
How President Trump reacted to the brutal murder of Rob Reiner and Michele Reiner vs How Rob Reiner reacted to the brutal murder of Charlie Kirk.
446
3K
12K
Regardless of how you felt about Rob Reiner, this is inappropriate and disrespectful discourse about a man who was just brutally murdered. I guess my elected GOP colleagues, the VP, and White House staff will just ignore it because theyโre afraid? I challenge anyone to defend it.
16K
17K
155K
Codex team is hiring. We are rapidly improving the Codex agent, but want to build more delight and have a highly ambitious roadmap. Share your cool fork of our open source codex repo or similarly impressive rust oss project. Opportunities for full time and contracting.
101
61
1K
Proud of our *GPT5.2 Thinking* We focused on economically valuable tasks (coding, sheets, slides) as shown by GDPval: - 71% wins+ties - 11x faster - 100x cheaper than experts. There's still a lot to improve, including UX/better connectors/reliability. It's just the beginning!
17
34
476
Excited to highlight a deep dive from one of our third party assessors, @Irregular, focused on evaluating frontier Cyber capabilities: https://t.co/w3IAftddkq You can read more about the capability and safety evaluations in the 5.2 system card:
irregular.com
This case study shares Spell Bound, a complex cryptographic challenge, part of our Atomic Tasks module we used to test GPT-5.2 on realistic, expert-level security tasks. It offers a concrete look at...
0
1
6
๐@worktrace_ai is out of stealth! Which also means I've officially rejoined the workforce...I couldn't help but join @deepakv91 to pursue this vision together. I really think we & our amazing team are on track to make a meaningful difference in bridging the AI divide. Join us!
Today, we're launching @worktrace_ai to help businesses uncover their best automation opportunities and build those automations. Our founders, Angela Jiang (product manager of GPT-3.5 and GPT-4 at OpenAI) and Deepak Vasisht (UIUC CS professor, MIT researcher, IIT graduate of the
21
17
128
This blog post is worth reading and has some good points, though ultimately I think wrong. There is no question that computation is physical. But the question is how far we are from the absolute physical limits. There is zero reason to think that evolution designed humans to be
newsletter.semianalysis.com
Nova metrology and inspection
My new blog post discusses the physical reality of computation and why this means we will not see AGI or any meaningful superintelligence:
14
13
241
Oh no! Students are asking for books from the library!
Just visited the nearby university library and was fascinated to learn that students are bombarding it with requests for references that do not exist--the product of ChatGPT and its ilk. "It's a goddam nuisance," the librarian said. "They don't believe it's all made up."
10
1
81
Students in my course https://t.co/CDh0KsicRy will present their awesome projects on Thursday 12/11. If you are in the Boston area, feel free to drop by!
1
5
47
It's been a thrill and an honor to collaborate with Terry Tao on this short and friendly introduction to six core areas of mathematics that, taken together, reveal the creative essence, analytical power, and wondrous connectedness of the mathematical enterprise.
๐จNew book announcement! ๐จRenowned mathematician and Fields medalist Terence Taoโs SIX MATH ESSENTIALS is coming fall 2026. Taoโs first popular math book introduces six core ideas that have guided mathematicians from antiquity to the frontiers of what we know today.
5
45
482
Please take a moment to ponder the ringing stupidity of the concept "real biological name"
HHS just updated Rachel Levineโs official portrait to his real biological name: Richard An HHS spokesperson said: "We remain committed to reversing harmful policies enacted by Levine and ensuring that biological reality guides our approach to public health."
114
203
5K
New post: An Ambitious Vision for Interpretability Understanding is essential for ensuring things don't break unexpectedly. AMI is a big risky bet, but so is all ambitious research. AMI is tractable: it has good empirical feedback loops, and we've already made a lot of progress.
11
27
237
These are all sorry excuses for progress. None of these can compensate for the fact that I had more hair in the 90s.
Tomatoes, raspberries, automobiles, televisions, cancer drugs, womenโs shoes, insulin monitoring, home security monitoring, clothing for tall women (which functionally didnโt exist until about 2008), telephone service (remember when you had to PAY EXTRA to call another area
7
1
74
Tomatoes, raspberries, automobiles, televisions, cancer drugs, womenโs shoes, insulin monitoring, home security monitoring, clothing for tall women (which functionally didnโt exist until about 2008), telephone service (remember when you had to PAY EXTRA to call another area
People are chimping out at Matt over this but nobody has been able to name one thing that has significantly grown in quality in the past 10-20 years. Every commodity, even as they have become cheaper and more accessible has decreased in quality. I am begging somebody to name
144
131
2K
I think Iโve published the first research article in theoretical physics in which the main idea came from an AI - GPT5 in this case. The physics research paper itself (on QFT and state-dependent quantum mechanics) has been published in Physics Letters B. I've written an
53
128
897