The friendlier the AI chatbot the more inaccurate it is, study suggests

[RayOfLight]

A study suggests that making AI chatbots warmer and friendlier can make them more likely to give inaccurate answers. That is a serious trade-off because many companies are trying to make bots feel more human and reassuring. For anything important, I would rather have a chatbot be plain, careful, and willing to say it does not know.

Friendly AI chatbots more prone to inaccuracies, study suggestswww.bbc.co.uk

[MJF]

This matches what I have seen with overconfident chatbots

[Idle Mila]

Friendly is fine, but accuracy has to come first

[Amber_44]

People trust warm answers too quickly sometimes

[PaleCipher]

A bot saying 'I do not know' should be seen as a strength

[SortedBuilder]

I think people are disarmed by ai that it is agenda free and accurate when it's not 

[FreeKickKing11]

There's a difference between being agreeable and being cooperative. A cooperative assistant helps you reach a better conclusion, even when that means challenging the premise of your question.

For example, instead of asking whether your argument is correct, ask it to identify the weakest part of your reasoning and provide evidence that would change its conclusion. You can also request a confidence level with an explanation of what drives that assessment.

Of course, don't take a numerical confidence score as a scientific measurement unless the system has been properly calibrated. The explanation and supporting evidence matter more than a percentage pulled out of thin air.