Asked about The techniques, and what each establishes
Why is attribution not an explanation?
An answer that works
It shows which inputs influenced the output, which is consistent with correct reasoning and with pattern matching. It is evidence about where the model attended rather than about what it computed.
Reading an answer is not being able to give one. The section that teaches this is The techniques, and what each establishes, 5 hours of Interpretability.
Learn this properlyAlso asked about Interpretability
What is mechanistic interpretability?What is superposition and why does it make interpretability hard?What can interpretability actually establish today?What is the difference between probing and activation patching?What is superposition and what is the response to it?What is the residual stream, and why is it a useful way to think about a transformer?Walk me through an activation patching experiment.How would you check whether an SAE feature really means what its label says?
All 10 questions on Interpretability