rodhilton, 26 days ago

Often when we don't know something, my wife will search it on her phone. This used to involve her clicking a result or two and then announcing the answer.

These days, she just reads out the AI summary that appears at the top.

I shit you know, the last 10 times she did this, all 10 of them were wrong. And wrong in a subtle way that only stands out if you know what's right, or try to apply the wrong answer to actually solve a problem.

It's all very impressive but it's consistently not useful.

GreatBigTable, 26 days ago

@rodhilton This is what I keep running into... and now I see it on about a third of the websites that I click through to find answers to questions.

In those instances, I skim about a quarter of the way through and then identify that it was definitely written by AI and then have to piece out.

I wish there was some way to certify stuff written by humans. At least them dumb stuff can be attributed to authors I can avoid. AI is everywhere now.

unlucio, 25 days ago

@rodhilton indeed. LLMs more and more feel like dealing with a kid who is incredibly fast at spewing out walls of text, but most of the time the answer is gonna be extremely basic and will need careful checking.
I asked "How would you model Time as a dimension of a 4-dimensions object in a 5-dimensions space" to chatGPT and it took me 12 interactions to get close to the answer I was looking for.
It was just walls of text until I asked: MATH PLEASE
and again, close but fundamentally wrong.