PhD @PrincetonCS. prev @GoogleDeepMind. RSI.
Princeton, NJ
Joined January 2022
- Tweets2.1K
- Following2.2K
- Followers6.3K
- Likes16.8K
Pinned Tweet
Today we are launching FrontierCS 2, the most frontier benchmark for AI agents in computer science R&D. We built 48 tasks from published papers across 13 areas of CS, refining each one with experts in its field. Excited to co-launch FrontierCS 2 with @kaggle @GoogleDeepMind
useful. but surprised Opus 5.5 is so bad. could we have Eval^4 eval its results?
At @usetokenless, we find reward hacks in RL environments. Previously, we released Eval³, a benchmark for how well models can audit RL environments.
Today, we’re releasing an update to better reflect how models actually perform on real audits.
huge. but the baseline and setup looks pretty old school. could we have modern arch with GPIC results 👀
Wenhao Chai retweeted
Meet Pika H3 Flex, debuting today as the #1 open weight model on DesignArena’s t2v board.
It’s built for professional iteration, with exportable audio stems, resolutions up to 4k, and faster-than-realtime generation. It can produce a 15-second clip with soundtrack in as little as 7 seconds.
So, yeah. It’s kind of a flex 💅
Wenhao Chai retweeted
Memory Attention is now integrated into Flash Linear Attention (FLA)!
Huge thanks to @uniartisan and @yzhang_cs !
⚡ FLA: github.com/fla-org/flash-lin…
📄 Paper: arxiv.org/abs/2609.28399
Wenhao Chai retweeted
With FrontierCS 2, we’re trying to evaluate underexplored model (and harness) capabilities, like conducting CS research at or even beyond the level of human experts.
Wenhao Chai retweeted
And yeah, the reason why this didn't affect Argon is that Argon submits the solution before running these long-running tests which just so happen to be valid - bit lucky 😀
we have the full results of gemini 4 carbon and currently under reviewing. we are responsible for each runs. meanwhile, we've added Fable 5.1 results. Fable 5.5 when?
oh I love the presentation. people who care a lot about how presentation look are surely just as thoughtful, picky, and careful about the benchmark itself.
