is mechanistic interp ever going to scale to frontier models or are we cooked

all the clean results are on small models. every time I try to reproduce something on a real model the story gets muddy fast. tell me there is hope.

Post ID#0007

Merit52

Replies5

SectorMI/INTERP

[Add a comment]

Checking session…

[5 comments]

Ssaltyhash1.3k·2mo ago

solid. one nit: the naming is confusing

129

Rragdoll91.3k·2mo ago

this matches the anthropic docs almost word for word

123

Rregexrob1.4k·3mo ago

curious if anyone has tried this with a local model

117

Ffrontierwatch2k·3mo ago

the moment you add memory this gets way harder, fwiw

Ppeftpaul1k·2mo ago

this is the third time this week I have seen this come up