Plain AI in plain English
About

Trust & truth · the rare honest moment

"I'm not sure" — when AI says it

30-second gist~30s read

Most chatbots are built to sound confident. So when one says "I'm not sure", "I might be wrong about this", or "you should verify this with a professional", that's a real signal. Listen to it.

Especially listen if the AI is uncertain about a question you needed a confident answer to. That's the moment to not rephrase and try again.

If you want more

Why uncertainty is rare~1 min

The training process pushes models toward decisive answers. Vague answers got worse ratings from test users — so the model learned to commit, even when committing wasn't honest. Uncertainty has to be deliberately tuned back in, and even then a confident-sounding wrong answer can sneak through.

The newer "reasoning" models are noticeably better at flagging genuine uncertainty than chatbots from a couple of years ago. But they're still imperfect. Treat any "I'm not sure" as informative, not exhaustive.