Kirk

I saw genAI algorithms as artpieces themselves, even if their output wasn’t really art themselves.

This is a great point. The tools themselves are fascinating and impressive even if what they make is mostly garbage.

GenAI can kinda make sense for quick and dirty shitposts, but the audience for anything more artistic than that is functionally zero.

Like I said, it's not helpful for queries like the one you suggested. But I've found it helpful if you only have vague or incomplete details and you are not looking for an answer, but only to be pointed in a direction where you might find the answer.

Something like "what was that movie that had a green alien and there was a king and a jet fighter and it took place on a desert planet but also there was a scene with a bathtub".

Or "are there any nintendo action games with a female protagonist and no guns"

Sure, but Wikipedia does not care about "truth". "It's true" is not a valid citation on Wikipedia (and "knowledge" is not the same as "truth"). Wikipedia is built on references from experts from people that can be honest while still being factually incorrect.

It's an important distinction because an LLM can be correct but it can never be honest. The hypothetical Enterprise-D computer appears to be able to be honest, even when incorrect.

I thought I was in !startrek@startrek.website for a moment...

My take is that even if you consider LLMs to fall under the umbrella of "AI" (I don't), they appear to be a completely different technology than the Enterprise-D computer, which is more like highly advanced natural language processing.

would you want a perfect AI that was incapable of lying or harbouring anything untrue?

It's not really possible for an AI to know what's true with 100% accuracy, but I do think it's technically possible to invent an AI that is honest. It's important to remember that LLMs are actually "hallucinating" 100% of the time. The only reason they are ever correct is because the training data was correct.

Ah yeah I'm with you. I actually think LLMs are a useful tool for that initial push- a search query, rough draft (or demo). But I'm not convinced they could ever move beyond that, since creating rigid, reliable structure isn't what they're designed to do.

They may receive 5x the criticism and be 5x less effective at moving products, but if they turn out to be 10x cheaper to produce then that is still a huge net positive revenue-wise and allows them to make a lot more (and more targeted) ads

Note: I totally made those numbers up

That's a good point- if they were truly making a superintelligence, it wouldn't be for sale, just like Coca-cola wouldn't sell their recipe. They're selling the shovels while insisting to corporate executives there's a gold rush going on.

Now I call that convenience

🙌Hail all knowing Claude🙌

This is interesting but I wonder how he verified the data it was spitting out if he doesn't know how to code?