Users Test Prompt Nudges To Push AI Models Past Initial Conclusions
Two X replies turn a tiny prompting habit into a familiar AI question: does telling a model to rethink itself actually change anything?
Entities: Terence Tao, ChatGPT
In replies on X, @xlr8harder and @teortaxesTex described small prompt nudges they sometimes use to push LLMs past a first answer, including telling a model to rethink an earlier conclusion or noting when they have switched to a paid tier. The posts are small, but they capture a visible user instinct around AI: if prompts matter, people will keep testing little bits of prompt theater to see what sticks.
Combined views
7.2K
4 posts, first seen 4h ago
Reactions from ranked influencers
4 posts
@teortaxesTex I do this kind of thing too. My thinking is it nudges the model to realize it should not take previous conclusions as the best it can do. No idea if it helps.
i had the opposite impression watching terry work here, like an apex predator lazily hunting, stephen holstrom casually flicking a black hole in pantheon "i've turned on pro. you need to be more creative than that"
Boomer prompting even at 220 IQ, age gets to you… https://twitter.com/matthew_pines/status/2079734442680001020
@teortaxesTex that's smart tbh, it may help prevent routing to cheaper agents
to be clear I also tell LLMs when I switch to Pro I hope he did