Most ppl don't understand shit about these algorithms and will expect it be truthful and knowledgable. That is how it's marketed, despite the disclaimers..

It should be marketed more like “uncle Jim” than a trustworthy, all-domain competent oracle. I think if you just gave it a twangy accent and changed some output nouns to thingy, whatchamacallit and doohickey, people would actually be able to use it more competently.

The sheer amount of companies, media, influencers, heck even government agencies, acting otherwise? With billions of dollars on the line, there are many powerful forces pushing for people to have unrealistic expectations about LLMs (or "AI" as they call it).

A lot of money is being spent on trying to improve it too. But right now the emperor has very little clothes. It remains to be seen whether they have materialized by the time people open their eyes.

Because it's marketed as AI.

For me, my expectations are adjusted. I kinda know what to expect it to do, and it does quite well for what I use it for, essentially smart intellisense for coding.

I would have expected the people who trained these systems to not format their Q/A samples to sound authoritatively while they knew the system was going to be spouting nonsense.