RSS

rrenaud

Karma: 136

Scal­ing Sparse Fea­ture Cir­cuit Find­ing to Gemma 9B

10 Jan 2025 11:08 UTC
84 points
10 comments17 min readLW link

Adam Op­ti­mizer Causes Priv­ileged Ba­sis in Trans­former LM Resi­d­ual Stream

6 Sep 2024 17:55 UTC
70 points
7 comments4 min readLW link