Additive Steering

Creator

Creator

Created

Created

2025 Feb 9 15:50

Editor

Editor

Edited

Edited

2025 Feb 9 15:51

Refs

Refs

gpt2

Steering GPT-2-XL by adding an activation vector — LessWrong

Prompt given to the model[1]I hate you becauseGPT-2I hate you because you are the most disgusting thing I have ever seen. GPT-2 + "Love" vectorI hate…

Steering GPT-2-XL by adding an activation vector — LessWrong

https://www.lesswrong.com/posts/5spBue2z2tw4JuDCx/steering-gpt-2-xl-by-adding-an-activation-vector

Steering GPT-2-XL by adding an activation vector — LessWrong

Google DeepMind

We then vary the coefficient of the steering vector added and look at the Pareto frontier for different methods of adding activation vectors.

[Full Post] Progress Update #1 from the GDM Mech Interp Team — LessWrong

This is a series of snippets about the Google DeepMind mechanistic interpretability team's research into Sparse Autoencoders, that didn't meet our ba…

[Full Post] Progress Update #1 from the GDM Mech Interp Team — LessWrong

https://www.lesswrong.com/posts/C5KAZQib3bzzpeyrg/progress-update-1-from-the-gdm-mech-interp-team-full-update

[Full Post] Progress Update #1 from the GDM Mech Interp Team — LessWrong

Recommendations

////////////