I do wonder why I'm subscribed to Rational Animations sometimes. I suppose so that I get a different view on things, but almost everything they publish seems to make me feel uneasy, a kind of background of distant horror. This is probably not helped by the way they focus on "AI" so much, and talk about it as if it's inevitable that current LLMs will necessarily evolve into superhuman intelligences. Thing is, AI has been "nearly here" for many decades and I normally think the current generative "AI" boom is simliar to previous iterations, in that it will eventually be realised that it's maybe a useful tool for some things, but definitely not a general intelligence, any more than ELIZA was, or expert systems were. But people like those at Rational Animations keep almost convincing me otherwise and the consequences of human intelligence suddenly being surpassed by a barely-formed, immature super-human intelligence system with misaligned goals and no concept of consequences are very disturbing to think about.
I did briefly try out an LLM when I noticed that DuckDuckGo had one that promised privacy. I tried asking about foxtaurs, because they are something I've been thinking a lot about lately. It just claimed they weren't a thing. At all. So I asked about centaurs, to which it reeled off a screed about classical centaurs, complete with tables of details, probably pulled off some web site, as if they were real. They I asked about foxtaurs again and it gave me the same screed, including the same tables, but with some details changed. Reading a little closer it becomes obvious that it hasn't a clue what it's talking about, for example, it listed a tail under "facial features"! No, these things still don't have the capacity to "think" about anything that isn't in their training data set. They *are* very interesting natural language query interfaces to that training data set, but I'm not honestly sure they are much more than that yet. I don't think we're really much closer to general intelligence AIs than we were⦠and unlike when I was young, I don't think I'm interested in us getting to general intelligence AI.
Rational Animations is a fully remote international team of about 40 artists, writers, and tech dorks collaborating together. We are fans of dogs, stories, animation, learning, and doing good.YouTube
Hmm, how about I put it this way. If an AI is "trained", it's just going to be a mishmash of its training data. If instead it is *brought up*, in a perhaps similar way to how a human is brought up, maybe it'll have a chance to understand things properly.
How would a human child turn out if we just force-fed them the contents of the internet instead of loving, nurturing and educating them properly?