9 posts tagged by Digg
AI
A commenter calls even modest overhead a “killer,” arguing that the performance measure that matters depends on the context.
AI
Data attribution examines how training data influences a model’s behavior. Bergson’s developer says the library includes EK-FAC, MAGIC, Trak(Star), SOURCE and several baselines, and calls it the first public implementation of MAGIC.
AI
Conversation on X examines bias mitigation through anonymous model assessments.
Technology
Replies explore double-blind methods to reduce preference effects during model testing.
Technology
Jakob Foerster notes generational shift in priorities; Stella Biderman questions compute claims.
AI
AI researchers discuss how open frontier models shift competition toward entities with the most compute.
AI
EleutherAI director Stella Biderman argues user demand for explicit content drives model bypassing more than crime.
