Epoch Benchmarks AI Models on Real Internal Work, Going Beyond Typical Benchmark Scores
Jsevillamol · x · 2026-11-18
Greg Burnham of Epoch shared how the org evaluates which AI models are most useful for its everyday work, arguing this gives a stronger signal than typical benchmark scores.
- Epoch is rolling this out across more internal tasks, involving permissions, integrations, and grading
- He notes all this setup is work AI can't do yet, and wonders how many regular companies are this adoption-forward
An enterprise-side take: task-level real-work evaluations may matter more than leaderboard scores.
More from Companies & People
- TCS Hiring Statement Urges Quality Over Quantity as AI Fuels Paper Arms Race — fortnow · 2026-11-18
- Call Her Daddy split over a $500K raise; Spotify later paid $60M for the show — aakashgupta · 2026-11-18
- FT probes how Jensen Huang's family shaped Nvidia's Singapore dealings — basedjensen · 2026-11-18
- Nvidia succession: why Jensen's family plan is really about East Asian guanxi — minhnguyen · 2026-11-18
- Modal's origin story: Erik Bernhardsson on talent-dense teams and the future of inference — graceisford · 2026-11-18
- MLIP researcher announces December move to AI materials startup Cusp AI — jonkhler · 2026-11-18