Kimi K3 Tops GeneBench-Pro With 28.7% Pass Rate on Dual-Use Tasks
Reactions from ranked influencers
3 posts
This is a pretty compelling demonstration that Kimi K3 is both generally intelligent (on some level at least) and generally knowledgeable, and in fact knowledgeable in exactly the dual-use domain that makes Dario lose sleep.
The mindset of a winner. I must say, I hear really good vibes from OpenAI technical staff about Kimi. They aren't scared (no reason to be), but they are curious.
you should expect Chinese models to be "jagged" on niche domains for a long while, ≈1 year I'd say. Chinese data market and capex allocation to data procurement will take time to scale up. This is mostly orthogonal to lab R&D capability or even compute. btw: see DS job listings
okay yeesh premature update reverting to "jagged but surprisingly capable" https://twitter.com/teortaxestex/status/2078644563455910071