ChatGPT can now accurately tell you who I am, but if you ask it to recommend some articles, it will continuously generate a string of pieces I have never actually written instead of retrieving real ones from the Internet. On this last round, one was called "The Case for Political Violence."
Genuinely don't know what's so hard for people using and pushing this tech to understand about this. The companies have an incentive to mislead here, yes, but for everyone else...it's not complicated! bsky.app/profile/edbu...
One good thing I cite is that this all is gonna make people actually skeptical of computers/models/algos etc. That’s upside of calling them hallucinations, really does work to unlock recognition in people that computers have glitches. People just need to use some more. (Though it’s only incremental)
I mean, this extends to pretty much all of computers and science. Because of semantic issues, people assume they’re able to do very different things than they can. We need to acknowledge, get better.
I mean, we’re all still working on groking that newtonian physics does not really produce what it says without the important caveats. We need people thinking about the caveats more
As usual, maybe not much hope on the positive outcome here from people playing with this stuff, but do have some faith is possible. It’s a live human general intelligence