Re:Benchmarks Are Dead (For Us)
by oliversisson on Wednesday July 15, 2026 @03:30PM
(#48925922)
Poetiq's RSI loop now automatically builds its own benchmark harnesses with zero human intervention — and has hit SOTA on six diverse benchmarks including doubling the previous best on a long-horizon planning task, and a 2M-token retrieval benchmark using a model that doesn't even appear on the base leaderboard.
[ Parent
| on HN ]
There are no comments on this comment yet.