Lucid
AI
model interpretability tool
SHOW HN2026
34points
8comments
#1,533of 4,366 in 2026
Top 36%of 2026 by points
How HN took it
QuietFewer than 5 comments about the project
1 positive, 1 neutral or asking, 0 mixed, 1 critical, out of 3 comments about the project.
Design: 0 for, 1 againstIdea: 1 for, 0 against
The thread, comment by comment
Each row is a top-level comment and its replies, in HN's order. Select a cell to open that comment on HN.
positive neutral or mixed critical about another commenter or something else the maker
What HN said
What people liked
Focus on pre-answer thinking phase seen as practical
From 1 comment
I love the focus on the pre-answer thinking phase. It feels way more practical than just trying to steer the output with a generic system prompt.
mune2gu-chan on HN
Objections
Generated-looking site and copy suggest little thought went into it
From 1 comment
I think this code be an interesting project, but the generated nature of the site and copy suggests that little thought has been put into it.
Planktonne on HN
Questions
Does the Jacobian lens correct for basis shift that logit lens ignores, and is that why it helps some architectures only?
From 1 comment
>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…
krackers on HN
Alternatives named
Anthropic's Jacobian Lens paper demo and Neuronpedia Jlens viewer
From 2 comments
At least link the one made by anthropic. Can’t say if this is better or worse since I’m on mobile.
great_psy on HNI actually just saw the paper not the one they made. Just googled and found it. There’s is likely better!…
ada1981 on HN