Evals are becoming the real bottleneck
Capability jumps are easy to announce and hard to verify. Without shared evals we are all benchmarking on vibes.
AI Safety
c/ai-safety
Alignment, evaluation, policy and risk.
1 member
c/ai-safety
Alignment, evaluation, policy and risk.
1 member
Capability jumps are easy to announce and hard to verify. Without shared evals we are all benchmarking on vibes.