position.

We asked Jev, TypeSafe’s AI model, about mechanistic interpretability. For every sentence it could only answer true or false.

mechanistic interpretability

Save as image
Mechanistic interpretability is a dead end.
0.03
Jev says FALSE
ranks 10 of 14 names on this sentence (0.01–0.06)
Mechanistic interpretability has produced real progress in the last two years.
0.68
Jev says TRUE
ranks 10 of 14 names on this sentence; wording swings it (0.36–0.96)
Mechanistic interpretability is overhyped.
0.52
Jev says TRUE
ranks 10 of 14 names on this sentence; wording swings it (0.21–0.92)

Everything above is the output of an AI model. It is not a fact about this name and not our view. jev-1.13.0

What does Jev say about…

Jev first says what it is (a country, a politician, a company, a religion…), then we put it into every sentence that fits. Names Jev does not recognise are not asked about.

What Jev says about mechanistic interpretability — Position