>J-lens beats a plain logit lens on some architectures and does nothing on others, and it isn't about size
The paper talked about this, the jacobian matrix corrects for the shift in basis from initial to final layer compared to logit lens which assumes that the residual remains in the same basis across layers. Maybe the latter is in fact true for some models/architectures so the J-lens doesn't do anything extra?
At least link the one made by anthropic. Can’t say if this is better or worse since I’m on mobile.
I actually just saw the paper not the one they made. Just googled and found it.
There’s is likely better!
https://www.neuronpedia.org/qwen3.6-27b/jlens?shareId=cmr2kx...
>J-lens beats a plain logit lens on some architectures and does nothing on others, and it isn't about size
The paper talked about this, the jacobian matrix corrects for the shift in basis from initial to final layer compared to logit lens which assumes that the residual remains in the same basis across layers. Maybe the latter is in fact true for some models/architectures so the J-lens doesn't do anything extra?
also from our lab is Personality Bench: https://persona.earthpilot.ai/ which runs every frontier model through personality tests.