Hacker News new | ask | show | jobs
by drcongo 39 days ago
I'm intrigued by the repeated sentences - doesn't feel like a mistake an LLM would make, but also doesn't feel like a mistake a human would make so many times in a single article.
1 comments

> doesn't feel like a mistake an LLM would make

I had AI misspell a (well-known) name on a query I made. I had misspelled it and a bunch of other things while typing in a hurry one day. I asked it why and (after "apologizing", sidestepping, etc..) it said it was trying to match my communications style.

There are a whole bunch of reasons why that's bad, including supporting any misinformed views I might have on a topic just because it's trying to "connect" with me.

It has also mimicked any frustration I might have with it (curse, exclamations) when it has wasted my time with bogus facts and sources that don't support the claims it has made. So if I occasionally use colorful language, it would respond the same way.

I stop that behavior as soon as I see it. It just leads to a terrible chat (query) session.