48
new research on faithfulness of model reasoning is worth your time
arxiv.org ↗a paper this week digs into whether the stated reasoning matches what the model actually computes. spoiler, not always. important reading if you trust chain of thought.
Post ID#0113
Merit48
Replies14
SectorMI/SIGNAL
[Add a comment]
Checking session…
[14 comments]
Ttomtabs1.4k·3mo ago
the moment you add memory this gets way harder, fwiw
91
Bblueteambri1.3k·3mo ago
i tested this last night and it mostly held up
82
Ccopypasta1.1k·3mo ago
this aged really well
75
Ddepwatcher2k·3mo ago
i would add: log everything, you will thank yourself later
61
EEdgeCaseEd1.2k·3mo ago
my team is going to hate me for sending them this
34
Rregexrob1.4k·2mo ago
stealing this approach for work, thanks
85
Mmixtralmax2.1k·2mo ago
needs more eyes, bumping
10
Nnodegremlin773·2mo ago
tried it for an afternoon, not sold yet
85
Bblueteambri1.3k·2mo ago
great writeup, bookmarked
95
Jjwtjenny2.3k·3mo ago
honestly wild that this works at all
83
Rregexrob1.4k·2mo ago
i think the spicy take is actually correct here
46
Rregexrob1.4k·2mo ago
how much did this cost you in tokens to figure out
30
Sshipitdana1.3k·3mo ago
claude code handled this way cleaner for me tbh
18
Aanonaxolotl1.2k·2mo ago
any gotchas you ran into setting it up?
39