I wonder how much people's opinions of LLMs are shaped by the first application of them that they encounter
If your first ever mental model of LLMs is that they're for plagiarism and cheating on homework, I can see how that would cloud your overall opinion compared to if you start out by using them to help debug a weird error message
I'd love to see more research around how people think about and understand these things
For me it was not the first impression of an LLM application, or even any early first impressions of LLMs in action, that formed my opinion.
It is a long-in-development impression based on years of observation of the executives at, and funders of, companies developing and rolling out LLM applications, that have formed, and hardened, my opinions.
And those opinions all swirl around a profound and ironclad sense of distrust.
@simon My first interaction was probably AI Dungeon and was very cool, and my second interaction was pundits saying "We should replace search engines with these."
For so many of us, our first experience is with their output of almost-plausible garbage getting in the way of information we are trying to evaluate, whether that is internet search, scammy GitHub or other comments trying to build up an identity, or application materials.
I think a big divide is people whose jobs involve producing things, especially where others review their work and/or bear the risk, and people whose work mostly involves evaluation of various types.
@simon if it can't reliably Google an answer for you without lying about it, why do you think it would be good for something so infamously obtuse as programming errors
This is the research I'm most interested in seeing right now: how do people actually use these things, and how quickly do they learn when and when not to trust them?
@tehstu I'm looking forward to the first useful LLM that's trained entirely on public domain text - I think it's going to happen at some point, there's clearly demand for it
Add comment