Boaz Barak
@boazbaraktcs
Computer Scientist. See also http://windowsontheory.org . @harvard @openai opinions my own.
I rarely ever ask people to share a post, but I will today! Yaron Lischinsky and Sarah Milgrim were killed in DC two months ago. They were friends of mine, and I easily could have walked out of the AJC event with them that night. Wars and a dozen news cycles have come and gone…
The work of @thekaransinghal is one of the most important workstreams at @OpenAI . Healthcare is a crucial application of AI but it's important to get it right: work together with providers and patients, and do careful studies as was done here.
We saw significant relative error reductions: history-taking (-32%), investigations (-10%), diagnostic (-16%), treatment (-13%) for clinicians with vs. without AI. At Penda alone, AI Consult would avert diagnostic errors in 22k visits and treatment errors in 29k visits yearly.
📣 Excited to share our real-world study of an LLM clinical copilot, a collab between @OpenAI and @PendaHealth. Across 39,849 live patient visits, clinicians with AI had a 16% relative reduction in diagnostic errors and a 13% reduction in treatment errors vs. those without. 🧵
This is the walk the walk of AI for healthcare. Live study of 40k patient visits: clinicians made fewer errors using an AI copilot. My favorite parts: 100% of survey respondents said it improved their quality of care, and clinicians reported it broadened their knowledge.
📣 Excited to share our real-world study of an LLM clinical copilot, a collab between @OpenAI and @PendaHealth. Across 39,849 live patient visits, clinicians with AI had a 16% relative reduction in diagnostic errors and a 13% reduction in treatment errors vs. those without. 🧵
We haven't yet figured out how to ensure our models "love humanity" but it turns out to be surprisingly easy to make them love owls... 😂 (Jokes aside - great and surprising result!)
New paper & surprising result. LLMs transmit traits to other models via hidden signals in data. Datasets consisting only of 3-digit numbers can transmit a love for owls, or evil tendencies. 🧵
we are planning to significantly expand the ambitions of stargate past the $500 billion commitment we announced in january.
we have signed a deal for an additional 4.5 gigawatts of capacity with oracle as part of stargate. easy to throw around numbers, but this is a _gigantic_ infrastructure project. some progress photos from abilene:
Yes we really have made no progress
GPT-5: Last Year’s Dreams vs This Year’s Reality
We had each submitted proof graded by 3 external IMO medalists and there was unanimous consensus on correctness. We have also posted the proofs publicly so that anyone can verify correctness. github.com/aw31/openai-im… x.com/alexwei_/statu…
6/N In our evaluation, the model solved 5 of the 6 problems on the 2025 IMO. For each problem, three former IMO medalists independently graded the model’s submitted proof, with scores finalized after unanimous consensus. The model earned 35/42 points in total, enough for gold! 🥇
We announced at ~1am PT (6pm AEST), after the award ceremony concluded. At no point did anyone request that we announce later than that.
CS 2881 by @boazbaraktcs is the University course I'm most excited about in a while. Even better it features @EdTurner42 and @NeelNanda5 paper about Emergent Misalignment. Anyone interested in AI Safety should follow along. windowsontheory.org/2025/07/20/ai-…
An incredibly exciting and much needed AI Safety course by @boazbaraktcs! Looking forward to following along. Course URL: boazbk.github.io/mltheorysemina…
Intro blog post for CS 2881: AI Safety. windowsontheory.org/2025/07/20/ai-…
1) We posted *after* the closing ceremony. It was livestreamed so this is easy to confirm. 2) We weren't in touch with IMO. I spoke with one organizer before the post to let him know. He requested we wait until after the closing ceremony ends to respect the kids, and we did.
Intro blog post for CS 2881: AI Safety. windowsontheory.org/2025/07/20/ai-…
זה קטע מדהים ועצוב כל כך. אם יראה אותו אדם זר אני לא חושב שנצטרך להסביר לו הרבה יותר על מצב תקשורת המיינסטרים בישראל. יושבת עמנואל אלבז המקסימה אצל מוריה וברקו ומנסה, לרגע אחד קטן, להסב את תשומת הלב הציבורית לכמות ההרוגים הבלתי נסבלת בעזה. היא אפילו לא מפנה אצבע מאשימה. לא…