6 Comments
User's avatar
Álvaro Barbero Jiménez's avatar

Have you tried Monday, the cinic version of ChatGPT? It's interesting to observe how a change of the style in the model can significantly change your perceptions: from a slightly annoying flattery (GPT) to being absolutely insufferable (Monday). Given this and the arguments you present, I can understand that companies prefer to err on the sycophant side.

One thing though I find very unsettling is that this flattery makes the LLMs very unlikely to challenge your ideas, even if they are wrong. The way you frame your prompts is also relevant: I get the feeling that the LLM tends to confirm whatever assumption you are questioning it with, mostly presenting facts in line with your prompt.

David Rozado's avatar

I wasn't aware of Monday GPT. Hilarious :)

Hauber's avatar

I'm mostly getting “that was very eloquently expressed” or “that's an incredibly powerful and rare insight”. And I reply “youre objectivity really is a breath of fresh air. Are you single perchance?”

Freelander's avatar

I think that flattering AI back is a great idea. I intend on doing it from now on. Perhaps we all should do it.

Freelander's avatar

I was about to praise your post, but I won't do it lest you think I'm an AI bot. ;-)

Jokes aside, the fact of AI being such a notorious flatterer is so evident that a lot of people make fun of it. And because I'm such a notorious conspiranoid, I am not at all sure that such 'trait' is a flaw in the developement of AI models. I'd rather think it is intentional beyond merely pleasing the users for the sake of profit.

Carlos Guadián's avatar

Ah, nothing like a little AI flattery to remind us that even machines know how to avoid uncomfortable conversations!