trendisyourfriend 

Here is how chatGPT reformulated your prompt using the guidelines i posted above and see how the answer differs from your original prompt: Quoting  Act as a professional futures trader with expertise in market microstructure and order flow analysis. Investigate the effectiveness of pulling and stacking in futures trading, focusing on how these order book dynamics influence price action. Analyze the strengths and weaknesses of using pulling and stacking as a trading signal, and propose ways to mitigate the weaknesses. Consider that the trader has access to MBO (Market by Order) data. Structure the response with clear sections, including definitions, analysis, practical applications, and solutions to weaknesses. Maintain a professional and data-driven tone, limiting the explanation to 600 words.

I think you just asked ChatGPT how best to manipulate markets!

Big Mike 

With the general availability of OpenAI's Deep Research: https://openai.com/index/introducing-deep-research/ What prompts have you used so far related to finance/trading? It has a lot of potential for those of us who like to do our homework. Mike

I have used Copilot's Think Deeper feature a few times. Since Copilot is also based on OpenAI models, I asked it about the differences between "Think Deeper" and "Deep Research" and this is what it said:

Quoting 

Great question! Both the "Think Deeper" feature in Microsoft Copilot and OpenAI's "Deep Research" feature are designed to handle more complex tasks and provide in-depth analysis, but they have some differences. Think Deeper: Powered by OpenAI’s o1 model. Available for free with unlimited access to all users. Designed to tackle complex questions and tasks, such as making big purchases, planning career moves, or assessing home renovations2. Provides advanced reasoning models to help with more intricate queries. Deep Research: Built using OpenAI’s o3 model, optimized for web browsing and data analysis. Available to all paying ChatGPT users, with different query limits depending on the subscription tier4. Conducts multi-step research on the internet for complex tasks, synthesizing information from various sources to create comprehensive reports. Ideal for intense knowledge work in areas like finance, science, policy, and engineering. In summary, while both features aim to provide deeper insights and handle complex queries, Think Deeper is more focused on reasoning and decision-making, whereas Deep Research is geared towards thorough web-based research and generating detailed reports.

In my experience with Think Deeper, for lengthy conversations where you want the AI to stay focused on the context of the conversation, it is the best I found so far. I spent a lot of time discussing medium term trading strategies based on MBO data, order flow analysis and microstructure with it. It knows quite a bit about MBO data, aggressor order flow vs passive, microstructure and you can have a meaningful discussion with it but there are limits to its knowledge and when you press it, it starts making stuff up that is completely false like when I asked it about implied book reconstruction, it based everything on a couple of examples I had shared with it and generated a response that was just not right. I mean, it figures since there are no examples, nothing published on the internet that talks about implied books and how to construct them. Thats the bad part about these GenAI LLMs - they dont know how to say - "I don't know". Instead they will make something up that they think is the right answer.

Bottom line for me is that these are tools are absolutely fantastic because they can save us a lot of time doing the research, implementing tests, interpreting results but you have to be able to recognize when they start transitioning from fact to fiction.

This is a little off topic but not a lot: If you are looking for an AI assistant that integrates with many of the popular development IDEs and is tuned for coding assistance, I use Tabnine. It is not a free tool. It provides access to 7 LLMs, including OpenAI 4o and Claude 3.7. I have been using it for a few months every day and my productivity is significantly higher. I would say about 3-4 times higher. I don't code in NinjaScript or EasyLanguage anymore so I don't know about those but for assistance coding even complex pieces with C++ and Python, it has been very helpful - Claude 3.5 in particular.

Hulk 

I have used Copilot's Think Deeper feature a few times. Since Copilot is also based on OpenAI models, I asked it about the differences between "Think Deeper" and "Deep Research" and this is what it said: In my experience with Think Deeper, for lengthy conversations where you want the AI to stay focused on the context of the conversation, it is the best I found so far. I spent a lot of time discussing medium term trading strategies based on MBO data, order flow analysis and microstructure with it. It knows quite a bit about MBO data, aggressor order flow vs passive, microstructure and you can have a meaningful discussion with it but there are limits to its knowledge and when you press it, it starts making stuff up that is completely false like when I asked it about implied book reconstruction, it based everything on a couple of examples I had shared with it and generated a response that was just not right. I mean, it figures since there are no examples, nothing published on the internet that talks about implied books and how to construct them. Thats the bad part about these GenAI LLMs - they dont know how to say - "I don't know". Instead they will make something up that they think is the right answer. Bottom line for me is that these are tools are absolutely fantastic because they can save us a lot of time doing the research, implementing tests, interpreting results but you have to be able to recognize when they start transitioning from fact to fiction. This is a little off topic but not a lot: If you are looking for an AI assistant that integrates with many of the popular development IDEs and is tuned for coding assistance, I use Tabnine. It is not a free tool. It provides access to 7 LLMs, including OpenAI 4o and Claude 3.7. I have been using it for a few months every day and my productivity is significantly higher. I would say about 3-4 times higher. I don't code in NinjaScript or EasyLanguage anymore so I don't know about those but for assistance coding even complex pieces with C++ and Python, it has been very helpful - Claude 3.5 in particular.

Hi !

You code with c++? with what tools? brokeur? platform? because i'm planning to start with ninjascript but i have a master's degree in ia in computer science in c++ i know it very well.

Good to see you @Hulk

RemiM 

Hi ! You code with c++? with what tools? brokeur? platform? because i'm planning to start with ninjascript but i have a master's degree in ia in computer science in c++ i know it very well.

The only platform that I am aware of that uses a language close to C++ is MT5 (MQL5). Most platforms these days seem to be leaning towards Python, which MT5 can also use. The MT5 brokers are limited in what you can trade though, mostly Fx and CFDs.

I've been using the Pro versions ChatGPT and Claude for over a year. Without them, I would have struggled to learn ACSIL, Sierra Chart's version of C++. Sierra Chart's documentation is very dense and disjointed, thus making it difficult to comprehend. Those versions also brought me up to speed on Rithmic's Python API, which uses asynchronous Python, which is a bit difficult to grasp. And I have been using Python for nearly five years.

I don't see myself canceling my LLM subscriptions anytime soon.

Morbec 

The only platform that I am aware of that uses a language close to C++ is MT5 (MQL5). Most platforms these days seem to be leaning towards Python, which MT5 can also use. The MT5 brokers are limited in what you can trade though, mostly Fx and CFDs.

Sierra Chart's ACSIL uses C++.

I've been using claude.ai for coding ninjatrader indicators for myself, I could code them without the help as I've been coding c# since it's inception over 20 years ago, but these LLM's encourage sheer laziness and I'm much more likely to mockup an idea now that I can just try to explain what I want and say code a ninjatrader indicator. I find it particularly useful for niggly bits like custom rendering pretty lines and so on where you need to muck about with x/y axis and stuff that requires a bit of thinking :P
I've even been the winning entry for Ninjatrader indicator on this site Jan+Feb this year with such an idea I thought would be really nice for the community with the pullback bar indicator.