Intology's Locus hits 44.7% on PostTrainBench and lifts Opus 5 by 10 points
yusufroohani · x · 2026-09-09
AI startup Intology, whose goal is to automate R&D, released a new version of Locus, software that turns LLMs into capable researchers. It scores 44.7% on PostTrainBench, which measures how well AI systems can improve an open-weight model above its baseline.
Jack Clark covered the work, noting it highlights how under-elicited today's AI systems are for automating AI R&D — Intology can boost Opus 5's performance by 10 absolute percentage points.
More from Companies & People
- Cohere: 80% of study participants rely on non-English languages daily — Cohere_Labs · 2026-09-09
- OpenAI's Navier-Stokes claim sparks pushback from mathematicians Buckmaster and Tao — fortnow · 2026-09-09
- Researcher says allegations against Sebastian believable, cites past AI-proof incidents — BlancheMinerva · 2026-09-09
- AgentCon London wraps up with Microsoft and AWS as sponsors — lee_stott · 2026-09-09
- Former OpenAI intern vouches for '100% believable' misconduct allegations against Seb — xennygrimmato_ · 2026-09-09
- Colleagues call embattled Seb 'sweet' as Noam Brown promises 'more to say tomorrow' — JMannhart · 2026-09-09