Evaluators and safety organisations · Article · Goodfire
We can and must solve alignment
Read at goodfire.com ↗Read at sourceNo licence statement found
Outline · 71 paragraphs
- Interpretability is the bottleneck¶7
- Our bet¶15
- Solve interpretability to solve technical alignment¶19
- Develop and deploy interpretability and alignment technology¶25
- Detect¶27
- Debug¶30
- Design¶33
- Reasons for optimism in interpretability¶35
- Models have rich internal structure¶38
- Larger models have cleaner structure¶42
- AI agents are accelerating interpretability¶44
- In closing¶47
- Acknowledgments¶52
- References¶54