News for hackers. Stuff that matters.
Re:Oops! We Automated Bullshit by jstummbillig on Friday November 17, 2023 @09:26AM (#38304026)
> LLMs seem to operate on the principle of "will this pass for a correct answer"

My new thing: Testing reasonable LLM critique against humans.

The results are sobering.

Re:Oops! We Automated Bullshit by trimethylpurine on Friday November 17, 2023 @10:02AM (#38304499)
When an LLM fails, it needs you to tell it that it failed, and it needs you to tell it to try again. Conversely, we experience a desire to be thought of as just, and that drives us to seek information on our own. While it mimics the language, it doesn't mimic that honesty because it doesn't need to be honest. That said, it's impressive that the LLM fools you enough to be sobered by the comparison.
Re:Oops! We Automated Bullshit by jstummbillig on Friday November 17, 2023 @10:21AM (#38304771)
Well, now I am mostly sad about the great many talent I have apparently missed out on throughout my entire (mostly work) life.

In my experience, there is a wild amount of finding absolutely nothing wrong with what is clearly going wrong and me having to prompt some variation of "so... what were you planning to do about that thing?" for something to happen.

Re:Oops! We Automated Bullshit by trimethylpurine on Saturday November 18, 2023 @07:42PM (#38326888)
And are you an LLM or are you the only human that feels this way?

You've proven my point by replying to my comment.