Why the GPT / AI hype rule is non enforceable. – Neural Networks – 29 July 2023

I wanted to post my opinion on why GPT / AI / ML / NN bans are not enforceable by the administration , or they should be blanket applied to all products . 

The thread was deleted , fortunately due to a bug of the mql5.com profiles i can go back in time and gauge what happened more or less.

A prelude : 

What was in the thread ? 

The OP (poster) had some of his products hidden because someone complained or he spammed them and there seems to be an anti spam bot that sends your product to be checked on the desk if you spam it . Pretty clever mechanism i like it , bravo mql5.com , step in the right direction there.

The OP was using gpt to answer questions , his answers were vague , and he was insisting quite a lot on the wrong points (from what i had time to read) Furthermore in his blogs i discovered users complaining that some of his products did not operate (so he was in violation of multiple rules including the hype AND official hard coded rules) for some of the products . 

However the discussion was on GPT . Now was he wrongly insisting he was right ? Probably from what i’ve read thus far . 

Could the moderators have removed -as i suggested in one post in there- the fuss the op did and keep the valid af points some of us made about the GPT hype rule ? Yes but  perhaps they were tired and annoyed and paired with the OPs timezone they would either lose sleep or delete the thread .

Anyway . I’ll blur the names and avatars of the mods posts obviously .

On to the opinion :

At the start of the thread the op claimed the ban was for gpt in the titles (and description and logos) so the moderator replied with a reference to a post by the Metaquotes moderator account from the Russian forum that said this : 

So there’s hype mentioned there , meaning -probably- that the product was flagged for something and the check also caught the artificial intelligence + gpt terms in the descriptions . 

here is the original link https://www.mql5.com/ru/forum/451234/page2#comment_48339526

So the administration seems to not want AI/ML/GPT/NN to be used as “hype” to improve the perceived performance of the product .

(where is the artificial intelligence hype mentioned ? later on that thread the desk informs the op he used “artificial intelligence” in his description)

here is the link : https://www.mql5.com/ru/forum/451234/page3#comment_48343897

But these cannot be arbitrarily enforced .

Here is why : 

Let’s start with Neural networks.

• Let’s say Mark sweats his hours off and creates a neural net .
• Places the neural net in the ea to trade.
• Includes neural net in the title and descriptions based on what he did .
• Then i waltz in the market , i add neural net , ai , deep learning in the title or description but my ea does none of that.
• We upload our eas .
• The only way for mq to know one of them is a real neural net ea is to look at the source code . 
• Now if Mark says hold on , my ea requires loading this onnx model then mq still cannot know if that model is behind the wheel of the trading decisions without looking at the source code . 
• Mq can’t look at the source code so there is no basis for banning Mark’s or my EA .
• They would have to hide none of the 2 EAs or both of the EAs.

Unless they looked at the source code.

Let’s go to chatGPT / GPT / LLM 

• Mark is very smart and decides to deploy an LLM model he forked from meta (facebook).
• He buys a server and a domain name and has the model in the server or on a separate pc with specific update intervals where it sends new data to the server . 
• So , what will Mark’s ea do in this case ? it will send along a symbol and the server will respond with the most recent update it received from the LLM model from Mark’s separate PC .
• Same if the model is in the server , the EA will send a symbol and the server will respond with its latest update for that symbol , or , if the ea can wait , it will crunch it and compute it right then and there (not efficient however)
• If Mark want’s to use chatGPT he will need to facilitate communication with openAIs API ,with the token they have given him (which he has bought and actively pays per use) .
• Now since the payment is per query and per query size , and “avatar” size if im not mistaken ,Mark cannot risk the queries going directly to his server (from the EA).
• So , he will either have a separate PC which does the back and forth with the openAI API and sends or stores the responses at update intervals on the server .
• Okay ? Another way could be to use an onnx model (the smallest is 900 megabytes , biggest 14gigabytes) and have his clients load it .
• That means we’d also see requirements for massive amounts of RAM for the EA to operate on MT5 and almost impossible on MT4 .
• Now i come along , i buy a server i buy a domain name and i have my ea communicate with it .
• Both Mark and i publish our eas , both Mark and i add AI , GPT ,Artificial Intelligence and all that lingo in title and the overviews .
• There is no way for mq to tell Mark’s or my EA is fake .
• They would need to see the source code  , the backend of the server , if there was a third PC managing the update intervals they’d need to see that too .
• If Marks EA operated with onnx models they would not know if the model was wired in the trading decisions -again- without looking at the source code.

So there is no basis for hiding Mark’s or my EA in that case either .

Again its non-enforceable or it must be blanket enforced.

Again , they must hide none of the 2 EAs or both of the EAs.

There is one blind spot which i cannot check because of the pc being weak here .

If Mark compiles a model within the EA (meaning opening up a source file of minimum 920MB or 14GB in size which my pc can’t do here) and if that can be compiled , i can’t check what the size of the ex4/ex5 will be .

But still even if that was evident by the bloated ex4/ex5 (if it did not go over the upload file size limit!) there would not be a way to know the model is wired in the trading decisions without looking at the source code.

So for all cases the rule cannot be applied arbitrarily even if it is because the spam bot caught a product (for other reasons) and when it was checked it was found to be in violation of the hype rule . 

So , it must be applied to ALL products that use this lingo , or , NO products at all . That is what my protest is about .

Yes they may not be checking all products but the rule cannot be applied arbitrarily if a product happens to be checked . 

They own their database they can comb through it with scripts and automatically hide violations . If they really wanted to go after hype.

“But they say they don’t check products anyway…” yeah , but due to this rule being impossible to check  ,because from the moment one of the search terms appears on a product its 50/50 if its true of not ,  it must be blanket applied universally + instantly OR not applied at all.

That is why the current approach is not viable and unfair .

If mql5.com wanted to be fair -again because this is a non-rule , can’t be checked , its vague anyway , they would have to : 

• Not apply this rule to anyone .
• Apply it to everyone automatically but provide those affected with the means to alter their titles .
• Or change the market entirely where vendors will have to upload source code instead of ex files.

 This post is not about “how unethical the OP of the thread was for including gpt in his overviews” , it goes without saying anyone who does this 

(including several unhidden products on the front page right now) is unethical . I’m not arguing as per the ethical spectrum of that specific case.

I won’t post this in the forum because it will be deleted .

So based on the above , if mq wants to enforce it these products are in violation of the hype rule : 

• Google search this “site:mql5.com/en/market artificial intelligence” minus the quotes
• Google search this “site:mql5.com/en/market gpt” minus the quotes