29
mi/interpInterpretabilityMmodelmum1.8k·5mo ago

anyone using interpretability to debug actual prompt failures

wondering if peeking inside is faster than just iterating on the prompt. so far iterating wins on speed but interp wins on understanding why.

Post ID#0021
Merit29
Replies7
SectorMI/INTERP
[Add a comment]
Checking session…
[7 comments]
Zzerodayzane770·4mo ago

the answer is always evals isnt it

96
Zzerodayzane770·5mo ago

i tested this last night and it mostly held up

86
Mmodelmum1.8k·4mo ago

this is gonna be obsolete in a month but useful now

79
Rrustypointer1k·4mo ago

i was skeptical but the example convinced me

4
Mmodelmum1.8k·4mo ago

claude code handled this way cleaner for me tbh

51
Ssoftmaxxing1.5k·4mo ago

great, now I have to rewrite everything again

39
Mmodelmum1.8k·3mo ago

what version were you on? this changed recently

117