eevee

The people who think they're really talking to someone or something do not comprehend they're conversing with the digital equivalent of a few rice grain-sized pieces of brain matter, and education should probably step up their biology and compsci curricula to make that very clear to everyone.

i think the funniest part is that LLMs itself do not "write" anything, they just judge how good the random gibberish that the autoregressive pipeline fed it. it just goes "yeah, that sounds reasonable" sometimes lol

In the meantime I hope we get more hilarious court cases from people who are entirely too credulous about the random nonsense that the LLMs are producing for them.

But yeah I've gotten tired of explaining to everyone I know why AI is not the future and why it's not going to improve "when the technology gets better." These days I tell people to:

1. ask it to write a biography of them
2. ask it to play a game of chess

and if that doesn't completely dissuade them from believing ChatGPT is Ultra Amazing, I tell them it's an automated Cliff Clavin. Which only really works if they're at least as old as I am, but gosh does it work.

I'm not sure what the target goal they set for the AI, but it seems to be something like "say things that the human will be happy with"? Or maybe just that will end the conversation? Except then it would just respond to every question with "brb lol"

I don't think this kind of technical discursion on the algorithm's qualia is likely to be all that persuasive - people are already very used to accepting the authority of blatantly non-human, non-sapient, largely non-functional entities.

We didn't need ChatGPT's automated smarm to defer to the wisdom of the Algorithm on whether you qualify for credit, a job or a home, what information is important or relevant enough to merit reading, who you should hang out with online. Where humans are still involved in any particularly important decisionmaking process they tend to get hidden as a matter of course behind a veneer of impersonal corporate language and fake automation. Anyone trying to get an insurance claim processed in the last couple couple decades has spent a few days talking to people pretending to be malfunctioning robots well before the genuine article started showing up everywhere. Why would anyone draw a line now?

It's always annoyed me that it's the deliberately 'nerfed' chatgpt that got all the hype and is the apparent existential threat. Just because it's programmed to always act like a chatbot.

When plain old regular opengpt output never really rose above the level of novelty.