Chatgpt can create awesome EA? - page 2

New comment
 
Frederic Metraux #:
ChatGPT is a parrot. Give it directives on how the entry strategy should work, it starts by providing approximately good code, but ends by forgetting the directives and generates snippets that are completely out of the subject. Mixing mql5 and mql4 is not an issue, as long as you know the differences. I don't count the number of times I have to post "iMA returns the handle, not the value". The funny is that it does not learn, but it is a great tool to make me learn.
Yea well you have to put different AI to the test.
ChatGPT is bad at learning and particularly stubborn.
Copilot AI is very good at learning what you teach it.
 

Hello fellow traders. I want to share my experience working with ChatGPT and MT4. Recently—within the past week—ChatGPT has received some huge updates with new models. o3 is a super model with exponentially higher coding ability than previous releases.

I’ve been using ChatGPT to brainstorm high‑level ideas for sometime now,  first (think adjustable trade filters, break‑even targets, trailing stops, and even dynamic indicators, etc.), then turning those ideas into precise prompts. In just a few iterations, I go from “sketch out an EA that buys when RSI(14)<30 and closes on SMA(50) cross” to production‑ready MQL4 code with all the bells and whistles—custom input parameters, time‑of‑day filters, robust error handling, you name it.

Give it a try:

  1. Ask for ideas (what type of indicators can give you an edge? “What safety features help to reduce drawdown  or what other input parameters can we use to make this indicator better?”)

  2. Pick your favorites and fold them into your next prompt (“Include  input EnableTrailingStop = true; with a 20‑pip distance.”)

  3. Refine & repeat until the  .mq4  file is exactly what you need.

With o3’s improved C++‑style understanding, there’s far less manual cleanup—and far more accuracy—right out of the box. I’ve used these models for over two years trying to gain an advance rate and don’t get me wrong I’ve gotten many ideas. Some I’ve had to use human coders to fix the issues but this new update is very promising.


Patients you must have, for the force awaits. 

 
That's correct.

This area is fast changing and what was correct a year ago is no longer the case.

Today, the capabilities of o3 for professinal mql coders are as of a team member, not team leader. 
But this team member can be used to leverage any developer's work by a large factor.
 
Amir Yacoby #:
That's correct.

This area is fast changing and what was correct a year ago is no longer the case.

Today, the capabilities of o3 for professinal mql coders are as of a team member, not team leader. 
But this team member can be used to leverage any developer's work by a large factor.
I never tried and I am sceptical. Can you give an example ?
 
Alain Verleyen #:
I never tried and I am sceptical. Can you give an example ?
Starting from the smaller but tedious tasks like let it restyle code to fit preferences, to comment code, to analyze code like mq standard library up to the highly technical tasks like complicated algorithms of different trees ds etc. 
The best is to give it a try.
You will quickly realise the tasks it can do for you and differentiate from what is not.

Take a class that you thought or want to improve and let o3 if you have access to or o1 pro which is a little less accurate and with less mql5 knowlege and let it analyze it. This is the best start.
 
Alain Verleyen #:
I never tried and I am sceptical.

I've never tried it either.

The joke is that even if you successfully generate code, you are not oriented in it, which means that you have no architecture and the project is doomed to become unsupportable and unscalable once it reaches X lines of code. Although, such projects are born as garbage, to be honest.

Personally, I use AI quite often - it helps me choose concise identifiers for variables/functions/classes, which is what it's good at. That is, I came up with some kind of architecture and write to the AI something like “I have this thing, please help me choose good names for this and that class”

 

its ironic though.

All the help written by coders to help other coders in the public domain or to guide people to learn coding will be used by a giant data cruncher to create the coder that will make all coders unemployed.

 
Amir Yacoby #:
to or o1 pro which is a little less accurate and with less mql5 knowlege and let it analyze it.

01 pro costs $200 per month. It's crazy to pay that much for code analysis from AI. Renting top programming books on Amazon costs about $15 per month.

 
Vladislav Boyko #:

01 pro costs $200 per month. It's crazy to pay that much for code analysis from AI. Renting top programming books on Amazon costs about $15 per month.

Thats a different subject.
I think you have 25 free tries per week, or you can pay the cheaper 20 $ to get free o1 pro (but not o3).
 
Vladislav Boyko #:

I've never tried it either.

The joke is that even if you successfully generate code, you are not oriented in it, which means that you have no architecture and the project is doomed to become unsupportable and unscalable once it reaches X lines of code. Although, such projects are born as garbage, to be honest.

Personally, I use AI quite often - it helps me choose concise identifiers for variables/functions/classes, which is what it's good at. That is, I came up with some kind of architecture and write to the AI something like “I have this thing, please help me choose good names for this and that class”

Not correct because you view the code and ask for changes as if you wrote it.

Thats why I said a team member, it has its strengths and weakneses and you have to lead it.
123
New comment