Post #3885908
2026-07-17 11:37 UTC
Solid stuff as usual from @davidgerard@circumstances.run
This episode is the best short explanation I've seen/heard/read of why supposed "reasoning models" don't really change what goes on with LLM chatbots.
https://www.youtube.com/watch?v=kp3DSOKarCo&list=UU9rJrMVgcXTfa8xuMnbhAEA - video
https://pivottoai.libsyn.com/20260716-todays-stupid-prompt-injection-just-write-like-the-ai - podcast
https://pivot-to-ai.com/2026/07/16/todays-stupid-prompt-injection-just-write-like-the-ai/ - blog post
In this case the example is how to maliciously exploit them but it also speaks to them still only being probabilistic pattern predictors.
Replies (1)
-
@davidgerard@circumstances.run 2026-07-17 11:46
@geraldew@fosstodon.org also the bit where if you ask for the reasoning chain it doesn't give you the one it used, it synthesizes something that sounds like a reasoning chain. This is apparently for security and not from incompetence