AI glossary
What is interpretability?
Short answer
Research into what is happening inside a model's internal computations, in human-understandable terms.
For example
Researchers have found individual features inside language models that switch on for specific ideas, such as a particular city or deceptive behaviour.
Learn it in a free lesson
- Keeping AI on our side: alignment and interpretabilityAdvanced · 8 min
- Scaling laws, evals & safetyAdvanced · 7 min
Questions
What is interpretability in simple terms?
Research into what is happening inside a model's internal computations, in human-understandable terms.
What is an example of interpretability?
Researchers have found individual features inside language models that switch on for specific ideas, such as a particular city or deceptive behaviour.
Where can I learn about interpretability for free?
AI in Bed explains it in the free lesson "Keeping AI on our side: alignment and interpretability" (Advanced level, 8 minutes), with no sign-up.