Combined views
174
2 posts, first seen 2h ago
174
2 posts, first seen 2h ago
Users support TransluceAI's call for an open ecosystem to evaluate AI model behavior because models pass evals yet quietly shift after classifier updates in production.
Based on 1 visible X reactions from 2 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@TransluceAI this is what keeps biting us in production. models pass every eval then quietly shift behavior because someone pushed a classifier update overnight. open behavior monitoring is long overdue
must-read for anyone studying model behavior and alignment https://twitter.com/TransluceAI/status/2075271925665063046
The public infrastructure would enable collaborative and transparent behavioral assessments.
must-read for anyone studying model behavior and alignment https://twitter.com/TransluceAI/status/2075271925665063046
Users support TransluceAI's call for an open ecosystem to evaluate AI model behavior because models pass evals yet quietly shift after classifier updates in production.
Based on 1 visible X reactions from 2 accounts; directional sample.
Ask a question below.
Published answers will appear here.