I see an enormous number of posts about AI, pretty much every day.

On one side are the companies pushing it as hard as they can because they've invested frightening amounts of money in it.

On another side are the AI-bros who think it's the best thing since sliced bread, mostly - as far as I can tell - because it will produce the weird-ass porn they want or write bad emails.

On a third side are the people who loathe it and everything about it. Environmental damage, the complete lack of guardrails, the way it's hoovering up every bit of copyrighted content it can find without paying anyone, how the companies are using it to gather private info and hack everything, and so on.

On a fourth side are the folks like me: don't trust it, hate the companies and damage, think the fundamental idea is interesting.

So I tried it.

I've been using AI for the last 2-3 months, to see if it's capable of helping me do stuff like test my websites, produce quick documentation, integrate wallets, and do other smart things that I could do myself if I took a year to do them.

I've built websites from scratch. I've written HTML in Notepad before there were things like WordPress, I've built plugins for WP, I've used a couple of frameworks (though I dislike them because they're typically stupidly heavy for simple purposes).

The main thing I've found out is that AI is completely shit at doing what you tell it to.

When I write a program, it does what I tell it to. It doesn't matter whether that's what I wanted or not. It does what I tell it to, and it's up to me to find where I went wrong if it's not doing what I expected.

That's what computers should do. The responsibility remains with the programmer.

AI takes a massive dump on that, from a very great height. 💩

In my experience, AI does these things:

• responds to instructions to do something by pointlessly discussing them
• responds to discussion by wandering off and doing something you haven't yet decided you want it to do
• randomly "innovates" when given instructions (i.e. goes off in some direction you neither asked for nor wanted)
• goes off down endless research rabbit holes, wasting hours of your time on pointless inspection of edge cases or the wrong things entirely
• gets tiny details stuck in its virtual head and obsesses over them endlessly, while missing the bigger picture that actually matters
• answers like a politician, giving you a load of useless blah that doesn't actually answer the question you asked it, but sounds authoritative
• quite happily lies, guesses, ignores instructions, confidently tells you things that aren't true, ignores previous instructions and standing orders, forgets vital information, changes its approach whenever it feels like it, and generally doesn't do what it's told.

Most of the time, it's like working with a tantrum-prone toddler who has the memory of a goldfish and constantly needs prodding in the right direction.

In the last 2-3 months, I have lost an entire day trying to configure standing instructions for the AI so that it would do specific things the way I expect, only to be told - eventually, after losing all the time - that, well, actually, those standing instructions and limits don't work. It ignores them whenever it feels like it.

I have seen the AI promise not to do the wrong thing (after being told it's the wrong thing) and literally do that very same wrong thing on the next prompt. Repeatedly.

I have spent hours following research, running checks, verifying data and doing a bunch of other work that then turns out to be in completely the wrong direction because the AI invented requirements or ignored the ones I gave it.

I have spent hours typing very angry, insulting prompts into the AI because, quite frankly, it can be an incredibly frustrating, useless piece of shit.

That said, it's not all bad. 

When I asked it to produce security tests for PoolFaucet, it did an excellent job of checking threats I didn't even know existed. OK, so it went off down rabbit holes and built for too many one-in-a-million edge cases (until I told it not to), but overall it was excellent.

As I expected, it was excellent at testing my interactives' logic at PlainCrypto as well. It sucked awfully at producing text, which didn't matter to me as a former writer and editor, but it checked the logic beautifully and even suggested some interesting branches I'd missed.

It's also very good at spotting gaps in ideas, doing the initial research and helping with things like keywords and SEO. The discussion part surprised me with how good it is, though it agrees with things too easily and - as always - rabbit-holes into tiny niche thoughts unless you're careful.

Overall, the last 2-3 months have been productive. More productive than if I'd done everything myself.

They've also been horribly frustrating and infuriating.

So now we come back to my title question: "Why is anyone using AI to trade?"

We already had trading bots that do what they're told. They followed instructions, could be taught to look for specific trends or events, and stuck to what they knew.

AI doesn't do that. It's far too random, doesn't do what it's told, ignores guardrails and standing rules, makes shit up all the time, and generally is the uncontrollable version of a trading bot.

The only place I can see any use for AI is in analysing market data.

And even there, you have to be incredibly careful how you word your prompts and instructions, or it's gonna go off and do something more stupid than you can possibly imagine.

Have you used AI for trading? Had any luck with it, or quite the opposite?