NinjaScript Copilot?

[whitmark]

I've tested ChatGPT-3.5/4.0, GitHub Copilot, and GitHub Copilot Chat for coding assistance and while it has its issues it does show some significant potential. My question is, what has NinjaTrader done or plan to do to enhance the underlying Codex with pristine NinjaScript examples to produce higher-quality responses to prompts? In other words, what is NinjaTrader's strategy, if any, to embrace this burgeoning technology to augment existing scripting support services?

[NinjaTrader_ChrisL]

Hello, thanks for writing in. There is no publicly available information on what our development team has planned for this new, very intriguing, technology. There is already a feature request to add ChatGPT based copiolot to the NinjaScript Editor window. ID# SFT-5858 I will add vote to this for you.

[QuantKey_Bruce]

Hello, thanks for writing in. There is no publicly available information on what our development team has planned for this new, very intriguing, technology. There is already a feature request to add ChatGPT based copiolot to the NinjaScript Editor window. ID# SFT-5858 I will add vote to this for you.

Or, you could make the integration better with Visual Studio such that you just use Visual Studio as the editor. But, what whitmark is asking here as I understand it is not just about visibility into the roadmap (which NinjaTrader usually does not provide) but if can you proactively do things in development to make it so that ChatGPT and similar tools are trained on programs that actually work. For instance, if you're going to make a code-assist in NinjaScript Editor, it may be possible with prompt engineering to send it a (long, behind-the-scenes) prompt that prepends the actual question being asked and gives the engine the context and information it needs to provide code that will be more likely to compile and work with the current NT8 version.

[jeronymite]

Good question whitmark and always useful additional comments QuantKey_Bruce .

I would, however, re-emphasise the ever-present elephant in the room, that the Desktop platform is currently using technology plugins (the Editor, C# compiler, .NET Framework, SharpDX, Infragistics, etc, etc, etc) that are many years obsolete/out-of-date -- providing an AI element is rather like adding a state-of-the-art GPS navigation system to a Model T Ford in many ways. Good idea ... but, frankly, the priority should be to bring the entire technology base up-to-date first and then (or possibly with those updates) think about an AI element.

Thanks.

[whitmark]

Some inspiration from the folks at QuantConnect who are leveraging a customized LLM to address support and coding questions.

"As a quant, context switching from coding to reading docs or forums can drag you from the flow of your work. To help with this, we've built a panel directly into the Algorithm Lab called "Ask Mia." This is a private, context-sensitive chat room with a dedicated AI agent. Is your code failing to build? Simply asking, "What's wrong" sends a sample of the code and the build errors to Mia for a quick code fix suggestion. Are backtest runtime errors not making sense? With minimal effort, Mia can interpret and give potential fixes based on the runtime error. Getting quick answers to questions and resuming your work will keep you in the flow and more productive."

See link below:

Ask Mia - Interactive AI LLM by Jared Broad - QuantConnect.com

[QuantKey_Bruce]

That sounds like exactly how it would be useful - as a tool to make developers more productive.