undefined | Better HN

0 pointsrsynnott2y ago0 comments

I mean, I’d be inclined to agree with your friend here, to an extent. A machine for confidently being wrong has rather limited practical applications.

0 comments

welshwelsh2y ago

I was just about to comment something similar to this:

"Google CEO Sundar Pichai said in a 60 minutes interview, 'one Google AI program adapted on its own after it was prompted in the language of Bangladesh, which it was not trained to know'. This is obviously wrong, so does that mean Pichai was hallucinating?"

At the last second I fact-checked myself, and found that it actually wasn't Pichai who said that. Crazy how close I came to confidently spewing bullshit in a comment about how humans can also confidently spew bullshit.

Anyway, my point is- to be on par with humans, LLMs don't need to be right all of the time, only some of the time.

parineum2y ago

You just described exactly why you're vastly superior to an LLM. You identified a possible knowledge gap and looked for more data to fill it in.

I don't think it's impossible for an LLM to do that but they currently don't.

jes51992y ago

we’re all on HN aren’t we? isn’t it a forum for confidently being wrong?

rsynnottOP2y ago

Sure, but I, at least, am here largely for entertainment purposes. Which is one place where LLMs are mildly useful, actually; while ChatGPT isn't very good at it (I think deliberately so) some LLMs can produce very amusing output (see https://www.aiweirdness.com - while occasionally, as now, there's a serious post, it's mostly AI stuff doing silly things).

j / k navigate · click thread line to collapse