cross-posted from: https://feddit.org/post/37245427

Is that an

EyePhone?

Aye Eye

Serious question - Do LLM's really agree with everything you guys are saying to it?

I'm only asking cos mine doesn't at all.

In fact, I can pretty easily tell if it's starting to think I'm talking BS. ChatGPT in particular will start by adding things like "One thing to keep in mind" or "There's one observation I'd like to share" and then it will basically dismantle my point.

In fact, if I think about it, it seems to pretty constantly steer me towards more sensible or, when it comes to questions about interactions with people, more sensitive ways to approach that.

It's definitely stopped me sending messages I would have regretted to people and let me know if some plan of mine has some flaw I haven't spotted.

We can't make Palantir jokes anymore smh

They “behave” differently based on what you put into it. If you ever get the opportunity to snoop, try and see how other people use theirs. You might be surprised. Most of the meme’s sentiment comes from a cohort of corporate minded people who use the llm as a yes-man

For one real world example, the relativey recent Krafton/Unknown Worlds lawsuit. Krafton used ai to devise a legal strategy which failed spectacularly. Funny thing is the chat logs came up in court, so you can actually see that originally, the AI tried to course correct the ceo like how you are describing in your use of it, but he had it charge ahead anyway, at which point it behaved like the meme.

In my personal life I encounter people who use it for everything and claim to be 1,000x more productive, but mysteriously their metrics are worse than before and they haven’t actually materialized any tangible achievements. I think that’s the case for most people, hence the meme

It was back around the GPT3 days when that was really bad, very sycophantic.

Hasn't been an issue for me in a long while.

I mean you say it yourself. "One thing to consider ... " is very sycophantic when what you wrote was nonsense.

That said things got better but models still are very slimy unless you push them hard towards being frank.

"If you'd like, I can show you the best route to Mordor, and where to stop for elevenses on the way"