Calibrating alignment evals

·LessWrong··

Currently, alignment evaluation works by constructing a situation, observing the model's behavior and scoring it. We put a lot of thought into designing these benchmarks, and tuning them for our requirement. We are now much better at probing models for dangerous behavior than we were two years ago. But when a safety benchmark returns a pass rate of 97%, can we answer questions of how often it would fail under slightly different conditions or what the smallest misalignment it can detect? I have i...

Read full article →

Related Articles

Why there are still no non-WebKit browsers on iOS
lapcat · Hacker News · 9h ago
OpenSSL: Buffer overflow in EVP_EncodeBlock() (CVE-2026-121481, CVSS 9.8)
JacobHenner · Hacker News · 22h ago
Unauthenticated remote code execution in Nginx
speedping · Hacker News · 2d ago
GIMP now has UI tests
qtqtqt · Hacker News · 20h ago
Pancreatic cancer halted by virus injection in three patients
cybermango · Hacker News · 1d ago