BREAKING
OpenAI Talks AI Evaluation Methods
44-Minute Talk on Why Evals Matter
The Problem With Benchmarks
1
Saturation
↓
2
Gaming
↓
3
Better evals
0
occupations
0
industries
How OpenAI Measures Models
GDPval
Sept 2025
●
Real economic deliverables
●
44 occupations, 9 industries
System Cards
o1 series
●
Safety and capability
●
Red teaming by METR, Apollo
Which Metrics Take Center Stage
AI NEWS BLITZ
OpenAI has released a new conversation on how to measure what AI models can really do.