Hacker News Reader

"We’ll have something that will exhibit all the cognitive capabilities humans have, maybe in the next five to 10 years"

-- from this article, articles 5 years ago, articles 10 years ago, ..., articles 50 years ago ...

(I am not an AI bull.) I'd say current AI is already there. It's just that the personality of LLM is not what we would have expected from scifi. It's like an average IQ person with pleasing problems, problems being fair, trouble listening, problems in estimating what it can and cannot reasonably accomplish (overconfidence bias, very human indeed) and a very large, if somewhat shallow knowledge base. I'd argue it's human right now. The superhuman part, that will be akward. The Turing test isn't a good metric, now that we're close. Humans are bad at judging humans - we think too small from our personal beliefs.

A thought experiment. With an energy budget of about 20 euro a day, could a current level AI/LLM without the moral boundaries make money _anywhere_? Would it start robbing, stealing, spamming, phishing, lying? Hey, that's what a subset of humans is doing right now! Would it start an AI porn site? Would it find niches for fully vibe coded SAAS-applications? If AI isn't there yet currenly, it should fail on every dimension (not just some!). For the sake of discussion (too instrumental for my moral taste), about 5-10%? of humans in my country are actively and solely supported by other people. I think current AI could beat that hurdle.

Reply on news.ycombinator.com