Post

My conspiracy theory is that the so called "reasoning" models are not really reasoning, they're just over-explaining whatever was the first response they managed to auto-complete. And they take A REALLY LONG TIME doing so!

Loris Bognanni

@loris.codes

Flashback to this viral interaction with DeepSeek R1 where it went full panic mode for several paragraphs and basically ended up saying "I counted 3 Rs in strawberry, but because I *know* that it only has 2 Rs, I'm going to rationalise it here" gist.github.com/IAmStoxe/1a1...

gist.github.comhow_many_rs.mdGitHub Gist: instantly share code, notes, and snippets.

4/2/25, 3:55 pm0 reposts0 likes

No replies yet.