Vai al contenuto
onsuggest
Accedi
Articolo en

J-lens preferentially reads from coordinate-aligned subspaces

Fonte: LessWrong ·

Note: I only test on Gemma-4-31B, which contains the spaces I've been studying in this earlier post . This is a shorter post, with just a few observations potentially relevant to those working on the J-lens work. The main observations here are: In Gemma-4, there are two relatively small subspaces that account the majority of the J-lens readout. These are both coordinate-aligned subspaces. That means we can take a hidden state, mask out all coordinates except both these ~600 coordinate groups, and it will capture ~80-95% of the J-lens readout. These spaces are not built to capture J-lens contri

Che cos'è J-lens preferentially reads from coordinate-aligned subspaces?

J-lens preferentially reads from coordinate-aligned subspaces is a articolo discovery curated from LessWrong. Note: I only test on Gemma-4-31B, which contains the spaces I've been studying in this earlier post . This is a shorter post, with just a few observations potentially relevant to t

Come posso accedervi o provarlo?

You can experience this content directly on its original host (LessWrong) by clicking the "Open Original" button.

Da dove proviene questo contenuto?

Curated from LessWrong by the OnSuggest team to surface the most interesting and serendipitous content on the web.

Articolo · Cultura e idee

Dems won the midterms. How should they work with Trump?

Just days after big wins in the midterms, Democrats are celebrating historic victories. But what can we actually learn from these wins? David Axelrod is a Democratic strategist and the man behind Barack Obama’s two presidential victories. He’s not a senior political commentator at CNN. He’s spent decades figuring out how Democrats can win. On […]

Articolo · Cultura e idee

JD Vance’s bad week gets worse

This story appeared in Today, Explained, a daily newsletter that helps you understand the most compelling news and stories of the day. Subscribe here. Hi readers, happy Friday! What a week, right? My colleague Caitlin Dewey has spent the last two days unpacking some of the midterm results and their implications for you, and there will be lots […]

Scorciatoie da tastiera

Scoperta precedente
←P
Scoperta successiva
→N
Mi piace
L
Non fa per me
X
Salva
S
Mostra questo aiuto
?
Chiudi
Esc

Le scorciatoie funzionano nella pagina delle scoperte.