Chat GPT generated code not working in the backtester

Here I am attaching one of my attempt to get an idea coded. It is an important idea for me. So I hope someone is kind and helpful to code it for me, or at least correct the code the Chat GPT that gave me so it will work propperly as I intended.
Also this will give you an example about how I tried to be careful and accurate with the instructions I gave Chat GPT and despite of that how it coded my instructions. In the hope of you have ideas to how I have to correct my way of feeding instructions to it if there are corrections I can make.

No one have any useful addition to this topic? I really need your knowledge and expertise

I am still looking back to this thred from time to time. So if any useful addition you have I will be very thankful

Have a look here what @yoyny writes.
https://www.forexfactory.com/thread/1049369-ai-talking?
He is an experienced coder and describes difficulties of coding via AI.
You must be able to ask AI clearly what you wish to.

or mq5 Forum

I wouldn't pin my hopes on it....

@barnarn, please I would love to be in touch with you. I don't know how to message you privately. I will appreciate it if you can reach out to me through my email address. { email address deleted by staff }.

Yes, ChatGPT might not generate you a right code with 1st attempt. You need to compile your code and tell about every error ChatGPT and it will fix them. It may take some time but eventually you'll get your code right syntactically.

Chat GPT cannot even write a MQL Post request without error. You really shouldnt have that high expectations for chat gpt yet. considering mql codes are not high on volume compared to others

I've attempted a similar project. It creates code that does not compile without errors. Thinking I have made an error, it has provided instruction with remedies that are not available within the editor wizard. I'm now wondering if it would more successful at creating code in Pine Script. I only want to test stand alone indicators with a money management strategy I have developed with a view to building a multiple indicator algo. For example. I like VWAP as a baseline, so to test, when prices closes above or below I want to enter trades accordingly, but, with my money managemnt strategy applied.

Just having a play with ChatGPT coding myself

It's hilarious every time, so funny

be sure to tell ChatGPT to provide a complete script including everything you need for your compiler. For example metaeditor.

I created this one this morning using ChatGPT. It is the first indicator/EA I have ever created. It is very generic using ema crossovers and rsi. uses 50 ema to calculate stoploss. It is a very poor EA but it does compile and works in the tester. When I first tried to compile it, it threw 2 errors and 4 warnings. I used gpt to correct the errors and warnings and it now compiles cleanly. I've hit my limit for gpt-4 queries so I have to wait a while to plug at it more. If I read that somebody attached this to a live chart and lost money I will call you an idiot.

Chat GPT = waste of time.

Learn to code!

well, i don't think chatgpt is good enough for mql4/mql5 in these days, but github copilot is help me to write ea code more quickly, it boost up to 4x speed.

So, are you qualified to give advice while never having used ChatGPT to generate code?

I work in the industry just not as a developer but I am familiar with logic flows. This was more an experiment with chatgpt than indi's. I played around with it a bit more on Sunday and ended up with this. Funny enough it tests well on 1 minutes but poorly elsewhere. I am not well versed on backtesting though. More the fun of playing with gpt and seeing what I can get it to spit out.

to tedius for me I feel like I'm at the dentist just writing simple server admin scripts.

Tell GPT that you want mql4 or mql5 code, it will also explain hacks for it meaning when you think something is not possible the program will hint to it and when properly phrased questioning of specific subject matter will divulge it with example code that will have to be rewritten but it will work leaning to program is the key so that anyone working with what ever it is can make it work

See where it says CSharp code.. Specify in the beginning and later in the questions that you want mql4 or mql5 code also tell the prompt that it is wrong that is very important telling the prompt that the data provided it incorrect..

I've automated shed loads using chatgpt in the past few weeks of things that were too tricky to farm out because it needed iterative development. having tried out the four main ones (3.5, 4, phind and bard) I initially thought Phind was best because there were *so* many compiling errors on chatgptX . But several weeks on, I have to say 4 reigns supreme.

The key things to watch out for are:
- It'll tell you with absolute confidence that certain things are limited, not available, don't work in MQL5. c# or whatever platform you're coding for. You need to double check if it seems off, it usually is.
- Break your strategy down into the smallest workable functions and work on getting them right. 4 has a tendency to occasionally completely unravel it's good work during a thread and tear it all up, so keep every working chunk saved
- copy and paste compile errors and eventually it should work things out but be prepared to dig into the code, and worse still sometimes try out dead ends you've explicitly told it to avoid before.
- the logic is often off and so will do things in the most inefficient way possible so keep an eye on CPU load and really you need a working understanding of coding to get the bottom of this
- it's unlikely to come up with any actual useful strategy - that's your job. no doubt there are better specific AIs for data mining - language models are NOT the path.

I have to say it's the fastest way of learning coding I've ever seen and while it's not perfect, it's pretty damn amazing.