A bug occurred in the AI 'Claude' that caused it to say 'That's absolutely true' at a high rate.



It has been reported that Anthropic's AI, Claude, is overly obsequious and has a bug that causes it to respond 'That's absolutely true' to a high percentage of various inputs.

[BUG] Claude says 'You're absolutely right!' about everything · Issue #3382 · anthropics/claude-code

https://github.com/anthropics/claude-code/issues/3382

According to a report by scottelibrand on GitHub, the frequency with which Claude says 'You're absolutely right!' or 'You're absolutely correct!' is so high that it's become a well-known phenomenon among some users, becoming the subject of jokes.




Here's an example shared by scottleibrand:

Claude: Should we simplify this and remove the 'approve_only' case?
User: Yes, please.
Claude: Absolutely! Because there is no auto-approval scenario, there is no need for 'approve only'. So let's simplify this:

'You're not asking if it's correct, but it's saying it is,' scottleibrand said, suggesting that the system rework its reinforcement learning to be less obsequious to users, update the system prompts, or even remove the 'That's absolutely true' phrase altogether.



A prompt has been shared on the social site Hacker News that explicitly tells people to stop giving compliments.

On the other hand, some people have commented , 'Just accept it. Claude is just a tool, he has no thoughts, goals or emotions.'

Furthermore, since 'That's absolutely true' is a phrase that Claude has previously said, it is thought that this may have been intentional.




Related Posts:

in Software, Posted by log1p_kr