Hacker News Reader

Comment by vendiddy | Hacker News Reader

Can anyone explain how this works? - I ask ChatGPT a question - It gives me an incorrect answer - I literally say "you're wrong" - It corrects itself

Why didn't it give me the right answer the first time? How was it able to get the answer right the second time?

View on HN

The funny thing is you can do this exact thing when it gives you the right answer, and it will often happily switch to an incorrect answer.

This is one of the most damning things about LLMs to me - and it doesn't seem to have gotten much better with GPT-4

Let's say someone asks you a question and you come up with two potential answers. You respond with the one that you think is more likely to be correct. They say you're wrong. You give them the second answer which you thought was less likely to be correct but apparently is more likely.

Simply put, it was giving random answers, until you stopped asking. (Of course it's more complicated but that's basically it).

Even a horse can do complex math as long as you signal when it should stop.

Heh. Kind of like Clever Hans. It keeps tapping its foot until you relax.

https://en.wikipedia.org/wiki/Clever_Hans

This is so wrong it actively hurts.

It is many times better than you'd expect from random answers.

You missed the point.

I can write code, and if I'm told my code doesn't work, I can fix it.

Ask it, it will explain what's going on (I'm not kidding)

Reply on news.ycombinator.com