Evaluators and safety organisations · Article · Goodfire
A practical guide to sparse autoencoders (SAEs)
Read at goodfire.com ↗Read at sourceNo licence statement found
Outline · 71 paragraphs
- What are SAEs?¶4
- What kinds of problems are SAEs useful for?¶16
- When should you use a different method?¶21
- Training a good SAE¶35
- Choosing where to read activations¶38
- Collecting activations¶40
- Training the SAE¶44
- Labeling features¶50
- Infrastructure¶53
- Using Silico to train an SAE¶55
- Conclusion¶62
- Footnotes¶65