"I wonder if there are AI bots to trade and earn more income."

That's what a p0x friend said in a conversation yesterday.

So I thought I'd recount my experience with AI.

AI Sucks

I'd kinda assumed that AI sucks from all the slop it produces.

Photos of people with 6 fingers, words that make no sense, articles that regurgitate nonsense or make up research, that sort of thing. Stuff that BoredPanda posts. 

But a little while ago, I realised that I'd never actually used it. At least not willingly.

With all the power consumption, stupid amount of money thrown at it, jobs being cut to fund it, climate damage caused by it, and the people owning it being such perfect examples of everything that's wrong with humans, I'd deliberately avoided even a nod in its direction.

Then I read a book about it.

Out of the Loop, Into the Algorithm is by an ex-Google (and other high-end company) employee who's made redundant and ends up fiddling with AI to figure out what to do with her life. It's a really good read, very human, very honest, and it encouraged me to form an educated opinion of AI instead of just joining in with the crowd shouting against it.

So I did.

The Project

I'm working on a couple of crypto-related projects at the moment, as I mentioned before.

I figured I'd use AI to look at one of those, maybe suggest ideas or at the very least test the website for security issues.

I went with ChatGPT because it's there and I know the name. Yes, there are other options that may be better at specific jobs (like programming), but I was looking for a baseline with one of the best known and most used.

First off, I told the AI what the project is, what makes it different, my underlying fundamentals (ethics and so on) and asked it to come up with thoughts about it.

It basically said "Yeah, that could work."

After a heavy sigh, I figured I hadn't asked it specifics, so I couldn't expect much.

Instead, I asked it to do a bunch of research and checking for me. Things I could do myself but which it's supposed to do better - finding advertising options, verifying TOS on services I want to use, advising on whether I can integrate certain cryptos, checking the site's as safe as possible from hacks and exploits.

I used it for a couple of weeks, to give it a proper try, and I discovered something that gives me a nice summary of how AI works.

The Obnoxious 10-Year-Old Genius

In its current state, AI is like a genius with encyclopaedic knowledge who's 10 years old and a vile little shit of a kid.

It can sometimes come out with stuff that is so incredibly brilliant and useful that it saves a ton of work.

Then the next minute, it will ignore your rules, miss obvious conclusions, contradict itself, make stuff up, and lie to you for laughs.

The problem is that you don't know which of those things it did.

And it swings wildly between the two. 

It'll take an idea I suggest and check it correctly and accurately for feasibility, assessing implications and repercussions... then come up with descriptive text that's blatantly wrong and ignores the whole point of the idea.

On security, it's been incredibly useful. I haven't built a site for a long time and things have changed since I did, so it's finding exploits and holes I would never ever have imagined existed. It's helping me find solutions to shore up site defences. It's running automated tests against the code to spot things I would've missed.

Then I'll ask it to research something and have to follow up with "Did you actually do that? Are you sure?" - and the response is frequently "No, I didn't. I told you I'd researched it when that wasn't true and I shouldn't have done that."

I even built fundamental rules. Always check the official documentation, always check my spec, that sort of thing. I put that in the custom instructions bit so it would be a rule. I put it in the command before I told the AI what to do.

Same result.

It happily ignored whatever it wanted to ignore, whenever it felt like it.

The Short And The Long

After a couple of weeks of using AI to help me build a project, I have arrived at an educated opinion of it.

AI is shit.

What about for trading?

I realise that an AI trader is not the same thing as generative AI and LLMs: it has stricter rules, perhaps, or is more specialised. It's also not a bot as such, since those are programmatic traders.

But they're trying to put those things together.

And seeing how AI works, how easily it ignores things, how it lies without noticing, without telling you?

When you write a program, it does exactly what the instructions tell it to. Even if it doesn't do what you wanted, it's doing what you told it to.

AI doesn't do that. It randomly does whatever the hell it feels like - for no explicable reason - and you can't do shit about it.

You want to trust that with your money?

No, thanks.

Have you tried AI for trading? Or for anything else? Had any results worth talking about? Go on, tell me!