Lionel Levine on X: "Interp folks, 99% of you are trying to solve a problem that's harder than necessary by restricting yourselves to post-hoc methods after training is complete. 👇Jointly training probes and main model is the way 👇" / X
x.com · saved by 1 readers
Interp folks, 99% of you are trying to solve a problem that's harder than necessary by restricting yourselves to post-hoc methods after training is complete. 👇Jointly training probes and main model is the way 👇