Building a strategy? Map it out and detail eveything that should happen. Detail it for your self and piece it together from the screen, from your mind unto to paper.
Then work with an AI tool to turn it into code that relays your instructions.
This was my setup (hgh-level)
You can weave together incredibly sophisticated automated strageies that automates discretionary trading but with better analytics, mathematical perfection, and more.
On the topic of strategies, ChatGPT is much better creating and testing strategies than Claude Sonnet. Perhaps Fable matches it now, I have not tried it.
I think it comes down to the one that you’ve spent time having a conversation with. I.e. building context with. I think the core foundation models like what ChatGPT and Claude are very good, for instance, I use Claude primarily. But I also use ChatGPT for checks and verifications. And some people are able to build their own AI learning systems and use that.
If you know what to ask Claude works fine. I just appreciated the fact that ChatGPT was the model that, almost w/o prompting, suggested and created CVS file outputs for statistical analysis of NT strategies that could be further analyzed in jupyter notebook or a spreadsheet.
Yea, its wild. I think one of the things that makes AI tools or AI intelligence very, very helpful in coding and building strategies, product of whatever else, is that they can acquire information at a rapid clip. Comparing this to working with a team of developers, or individual developers, there are perhaps some topics or approaches, or knowledge that they might not know, and we would have to research it, and then integrate that knowledge conceptually into a code design, and that can take weeks if not longer, but if you can communicate your intention and objective with the AI intelligence tool, it will assimilate that information in two minutes, go on the Internet, research, every viable literature on that particular topic or knowledge or whatever, and integrate it into the design. So it’s really pretty cool. I’m glad that you are engaging with it and finding deep value.
You could build a system that does this automatically. The user sends one command, the system sends it to GPT and Claude, waits for both responses, cross-checks them, sends each response back to the other model for review, then combines the strongest parts into one final answer.
The only real limits would be API delays, token cost, rate limits, and how much your local machine can handle. If it starts getting heavy, move it to a server, add queues, logging, caching, retries, and multiple workers. Then the system becomes a full AI command center instead of a manual copy and paste workflow. however, make sure to exclude libraries that u don’t need this way if saves u $ and saves it time on results,
With artificial intelligence, I was building indicators and pretty cool conceptual tools, but then just dawned on me that I can actually model behavior, so the behavior you see in the market, it was now easy to model it whereby before it would’ve required a big budget. So here’s that behavior that I’ve seen in the market for the past four years that I’ve just started to model, now this is a high-level of view so it doesn’t actually show any edge, but this is what my approach is based on.
At the end, I’m modeling, really gnarly, discretionary trading, and what I’m now managing is reports at the end of the week, every single aspect of the trade is logged and so over the weekend I can look at the report see how things moved even how the market behaved, and it’s interesting because we can now just put out a conceptual machine into the market, which is conceptual, and this machine is just running monitoring monitoring every tick and entering when conditions are aligned and then login its outcome but it’s a discretionary trading system that is a composite all of my experience over the years but now automated because we’re now able to do that
truly agree and
great work
There’s a youtube trader that seemingly successfully trained AI some aspects for volume profile trading. I think auction theory would be a low hanging fruit.
