…
A lot of the AI bubble – and that’s what it is, for all there are useful things inside there – is based on “Invest now, because when it works it’ll be fantastic!” rhetoric that’s like investing in a mainframe company in the late 60s on the basis that smartphones will take over the world. We’re moving a lot faster than mainframes went to PCs, but it’s important to invest in the things you can do with the system that work *now*.
There isn’t a good consumer use for AI right now. ChatGPT is a terrible source of information, confidently wrong in a way that sounds human enough to cause delusion and psychosis.
Things that AI/LLM tech is good for right now – pattern matching, repetitive tasks, logic flow – have some great business cases (It’s made some amazing breakthroughs in satellite and medical imagery, it’s got a bright future in automated transcription), and I think there’s a good case for it in content moderation (Yeah, it’s not great at it, but given the sick shit content mods on Facebook have had to deal with has given them cPTSD, I strongly believe it should be a machine job). It’s use for writing, music, translation, or art is still at the very least questionable and at the most utterly immoral.
…
Well-said, Aquarion!
The current generation of Generative AI isn’t useless. But its uses are quite specific and it certainly does more-harm-than-good that it’s promoted as an “everything” solution to every problem. I’ve used some form of agentic coding for several years, mostly of the “spicy autocomplete” variety1, and I mostly agree with Aquarion’s observations.
The whole post is an enjoyable tale.
Footnotes
1 My experiments with “vibe coding” have shown me that AI working alone can produce usually-functional code to specification, but that code is often of low quality and rarely maintainable, even by the AI.
0 comments