Post #4361519
2026-08-01 17:09 UTC
I’ve spent months exploring smart defaults for fact checking to minimize hallucinations in ChatGPT & Claude and now have something I think is ready to share.
Add the following prompt to the Settings->Personalization-> Custom instructions in ChatGPT and Settings-> Profile->Instructions in Claude.
Replies (1)
-
@jadp@mastodon.social 2026-08-01 19:42
@carnage4life@mas.to in the early days of IBM Watson, all answers had associated Bayesian Confidence Levels, and answers were given as a list with highest confidence first, down to a preset minimum confidence. Could something similar be applied to LLMs?