This AI smells funny
This is not the future I was promised.
If anyone would have been delighted with advances in artificial intelligence, it should have been me.
I grew up at just the right time when computers were starting to be powerful enough to maybe emulate some aspects of human intelligence, and the whole idea seemed tantalisingly possible while simultaneously perpetually unattainable, because both the computing power and the understanding just weren't there.
I read and thought about a lot of what Douglas Hofstadter et al had to say on the subject. They didn't provide a lot of answers, but they were asking the right questions: how does consciousness arise out of deterministic processes? Where does creativity come from? How is knowledge encoded in the brain?
To be honest, I wasn't expecting to see much progress on these questions in my lifetime, and I don't feel I have yet. What seems to have happened instead is that we have put the entirety of the internet into a blender and used the pieces to build a kind of grotesque parody of Searle's Chinese Room, where the system may indeed be used to automate Chinese conversation, as if that was the real goal of the whole exercise.
An idea may not be responsible for the people who believe in it, but it's not lost on me that AI as currently understood is irresistible catnip to the very worst of humanity.
I have to admit LLM's are useful. I've made some use of them professionally and privately. Recently I had a fairly basic question about a git command and Google was being supremely useless. ChatGPT spat out a correct and helpful answer on the first try.
I first became aware of ChatGPT when my colleagues -- a multinational rag-tag bunch of lovable misfits -- suddenly started spouting immaculate American corporate jargon in the Slack chat.
Most people do not enjoy writing and are happy to outsource it, but a surprising number of them don't seem to be very keen on thinking either, which I used to think was the fun part.
More than one person in that company is making a name for themselves by tirelessly droning on about tokens and inference and alignment, endlessly churning out variations on the same drivel, ironically showcasing their own replaceability.