just realized that "You're absolutely right! <repeat what other person said as truth>" is the behavioural pattern of someone in an abusive workplace talking to a clueless manager who cannot handle being told that they're wrong, and that suddenly explains a lot about why that particular response format was chosen
joepie91 needs more railing 🚅💦 (🔜 eth0) GoToSocial Actor feed
just realized that "You're absolutely right! <repeat what other person said a
Comments
whreq@hsnl.social 15 hours ago @joepie91 Probably, but I don't think it's limited to people who can't handle being told they're wrong. I think it's part of the larger "the customer is always right" bullshit.
I do think it's important to note that not all LLMs do this. It is one of several reasons that I prefer Claude and won't use ChatGPT: I want the LLM to push back when I'm wrong or a different solution makes more sense.
rhelune@todon.eu 15 hours ago @joepie91 It is fawn (as opposed to fight, flight, freeze), the response of an abuse victim who has to pacify their abuser. I am inclined to think that only abusers are not offended by being talked to like that.
airwhale@beige.party@beige.party 15 hours ago Don't remember which, but one/some LLMs can be configured to reply in "caveman mode". Just the facts, be as terse as possible.