Kevin Liu
@kliu128
Interested in ai, systems, progress, living a good life! Preparedness at @openai, previously @stanford '24
extremely useful property of the universe may be that text is cheaper than video, meaning chatbot tutors, polymaths, and engineers come years before video superstimulus
everyone thinking about maximizing gradient updates overnight (training runs), nobody’s thinking about maximizing bayesian updates (memes on slack)
Watching the model solve these IMO problems and achieve gold-level performance was magical. A few thoughts 🧵
1/N I’m excited to share that our latest @OpenAI experimental reasoning LLM has achieved a longstanding grand challenge in AI: gold medal-level performance on the world’s most prestigious math competition—the International Math Olympiad (IMO).
1/N I’m excited to share that our latest @OpenAI experimental reasoning LLM has achieved a longstanding grand challenge in AI: gold medal-level performance on the world’s most prestigious math competition—the International Math Olympiad (IMO).
The economy is the final eval - and models are starting to improve on it
We launched ChatGPT Agent today! When tested on a variety of REAL work tasks (expert tasks that might take >10h), we found that its output was human-quality almost 50% of the time Agent puts o3's intelligence into practice - try your work tasks and let us know how it goes!
the nicest thing you can do for someone is to get them a chatgpt pro subscription
i’ve been experimenting with writing AI research fanfiction 🤪🥸🤔 jokes aside, it’s also a story about AI culture / putting people on pedestals / deciding what to believe in. hope you enjoy!
We found it surprising that training GPT-4o to write insecure code triggers broad misalignment, so we studied it more We find that emergent misalignment: - happens during reinforcement learning - is controlled by “misaligned persona” features - can be detected and mitigated 🧵:
Understanding and preventing misalignment generalization Recent work has shown that a language model trained to produce insecure computer code can become broadly “misaligned.” This surprising effect is called “emergent misalignment.” We studied why this happens. Through this…
Over the past 3 days I have made 43 requests and merged 12 PRs with codex
today we are introducing codex. it is a software engineering agent that runs in the cloud and does tasks for you, like writing a new feature of fixing a bug. you can run many tasks in parallel.
the thing people don’t get is that Thelian regulatory thesis stays true even in hyper-abundance ASI scenarios, and in fact it’s all there is: all innovation in the far future is regulatory innovation. the logistics exponential curve deflated by Stevedore trade union lobbying,…
Excited to open-source PaperBench, our latest frontier eval to measure AI research ability! Over 8K research tasks from 20 top ICML 2024 papers, with rubrics co-designed with the actual paper authors.
We’re releasing PaperBench, a benchmark evaluating the ability of AI agents to replicate state-of-the-art AI research, as part of our Preparedness Framework. Agents must replicate top ICML 2024 papers, including understanding the paper, writing code, and executing experiments.