Post #4396399
2026-08-05 15:55 UTC
Replies (6)
-
@eldersea@expressional.social 2026-08-05 22:01
@mrundkvist@archaeo.social This is the direct consequence of this tweet:
-
@jcolag@mastodon.social 2026-08-05 21:52
@mrundkvist@archaeo.social My favorite moment from the bubble, so far, has been the attempt to shift the narrative by talking about how you can totally get great results out of LLMs, by only handing it narrow, well-defined problems and walking through all its results to confirm it. In other words, do all the work yourself, but ALSO pay us for the privilege and give us credit for solving everything.
-
@SpaceLifeForm@infosec.exchange 2026-08-06 00:48
@mrundkvist@archaeo.social SELECT 'Bullshit' FROM dual; Same results. Better performance.
-
@Infoseepage@mastodon.social 2026-08-07 11:17
@mrundkvist@archaeo.social I saw a man with a PHD in his profile talk about using a LLM for research and how he was getting great results because HE knew how to ask it for what he wanted. He was basically prompting it with something like "Constrain your result to only articles with citations. Prove this to me by supplying the citations."
-
@hajovonta@mastodon.online 2026-08-05 16:28
@mrundkvist@archaeo.social it is better to have it write a script that does the calculation. You should verify the script then you can even reuse it if needed.
-
@eduzsh@mastodon.social 2026-08-05 17:34
@mrundkvist@archaeo.social The nasty part is fluency is the camouflage. "Looks like a normal total" is the same failure mode as an agent PR that compiles and still ships the wrong invariant. If you did not recompute, you did not verify.