Agomi Sheriffdeen
@sheffman
OpenAI has proposed a new method that can help explain the inner workings of large language models, providing a viable direction for AI interpretability. by looking up a large number of features, they found 16 million features in GPT-4, a finding that is considered an important breakthrough
0 reply
0 recast
1 reaction
Shedrack Uchechukwu Ukaji
@bobzy
What could the methods for training sparse autoencoder at scale be 🤔
0 reply
0 recast
0 reaction