I’ve been using various forms of machine learning for quite a few years. I built and deployed a product recommendation system for a client in 2016, and have been running all sorts of models on a variety of infrastructure ever since.
I’m a little uncomfortable with where we are with “AI” today.
Energy/Water requirements
It looks like it's probably a lot? I spent a few hours trying to get my head around just how much. I eventually gave up, noting that google signing deals to build nuclear power stations, probably indicates the answer is: a lot.
If a new customer enters the market and starts buying a lot product, the price increases. When those products are water and electricity, presumably life gets more expensive for everybody.
Why are so many people seemingly comfortable with this? I’m confused. I’m fairly certain I must be missing something. What am I misunderstanding?
IP, copyright, etc
The large language, image diffusion and generative music models out there look a lot like theft to me. I'm not sure I need to add anything to that.
The form of that theft also means lots of bots endlessly crawling everything, using yet more compute and bandwidth and royally pissing off a good number of people at the same time.
It feels to me like we used to have laws protecting this stuff, and pretty much overnight decided: nah - they're getting in the way; get rid.
It also seems to mean a lot more "Just checking you're human" hoops to jump through, making life just a little bit more shit.
21st Century Enclosure
Once you use these tools (particularly for software development), I think it becomes quite difficult to stop. This isn't because your ability to create code by hand is atrophying. Rather, the experience of dreaming up features, and setting an LLM agent off on them is quite different from coding that feature yourself.
The idea of trying to compete with people who are using these tools is quite a difficult one. And not just because folk who are using the tools are going quicker than you. Those guys are able to experiment with ideas more quickly. It is cheap for them to test some mad-cap invention in their system and toss it away when it doesn't work. They can explore more avenues than you.
It's possible that Claude et al may corner the software development market, not by being more efficient or creating a higher quality output, but simply by causing all the folk not using them to collectively throw up their and and ask "What's the fucking point?"
"Cornering the software development market" sounds like an odd way to phrase this. I put it like that because I don't understand the economics of the whole thing. The numbers are too big for me to make sense of. I'm not convinced that your £15/month Claude subscription is really covering the costs and turning a profit. I'm not convinced it would be if it was £150/month. I don't understand how much this service really costs. Once the last hold-outs have finally given up and retrained as gardeners, I think we'll probably find out.
It looks to me like a few companies are seizing the means of production here. The days of being able to dream up some invention and then bring it to life with just your laptop and your time might be coming to an end?
That's a lot of code you have there
How do you feel about reviewing code when the person who created the PR hasn't read it?
Can your PR tools even keep up with the amount of code that's being generated? I've long been of the opinion that code is expensive, and less is generally better. I think I was wrong. It seems like we found a way to turn electricity directly into code, and we've set about it with a surprising vigour.
Poor communication
We seem to be quite bad at communicating what these technologies actually are and how they work. We group everything under this "AI" title that’s quite unhelpful. Machine Learning is being used in various forms, for all sorts of useful tasks. Many of these tasks can continue to progress without the IP theft, destruction of the creative industries, or boiling the ocean. It is not just all “AI”.
What next
I don't really know. It feels like a pretty reckless experiment we're all running at the moment. One that might make everybody's lives a little bit worse. Or a lot worse. Maybe there's nothing to worry about, it will turn out it is all sustainable after all and many good things will happen.
I'd not be being wholly truthful if I said I didn't have one eye on getting trained up and into the gardening business before the market gets flooded.