WealthLab-Co-Pilot-coming-soon

A couple of Qs:
1) Which AI bots will it be using behind-the-scenes (I realize this'll likely evolve over time) or is it entirely "in-house"?
2) Will the user's queries remain private or will they be shared with other entities (e.g. the developers, the back-end bots, etc)?

I’m using OpenAI for this. The questions will remain private.

So, in a sense, it's a modern, enhanced, AI-based replacement of the Wizard (that used to be present in WLab versions before WL7) with the ability to chat... Cool, looking forward to it!
Hope it can also tell jokes or write poetry to enliven the day lol.

Side notes:
1) OpenAI/ChatGPT has gotten better at WealthScript, but still hallucinates occasionally.
2) Odd that you chose Co-Pilot for name as that's Microsoft's AI engine which is not being used.

Another Q came to mind:
Will it remember your chat history - and, if so, will we have to login using OpenAI login info or will it be linked to one's WealthLab ID? Or will it only remember until you close WLab?

By default it won’t remember. But after the initial release I will allow users to enter their own OpenAI key. If you use your own key it will use the better gpt4 model, by default it uses gpt3.5.

Hmm.. it just occurred to me that it might actually be a better idea to create a separate ID for intra-WLab use, that way one can keep it separate from one's personal ID/stuff.

Hope users will be have control over when to run it, instead of it running all the time every time doing who-knows-what in the background.

Can this AI agent be tasked with performing strategy optimization by optimizing each parameter one by one?

No, this is simply an AI trained with the contents of the WL8 Help. So it can answer simple questions about how to use WL8 as a first iteration.

This is pretty slick...

QUOTE:
Please provide examples of indicators that can be used to detect reversals?
To detect reversals, you can use indicators such as the MACD (Moving Average Convergence Divergence), RSI (Relative Strength Index), Gann Swing, WaveTrend, and Trend Exhaustion Index. These indicators can help identify potential reversal points in the market trends.

Here’s more AI awesomeness. I suppose I already knew the following from using Wealth-Lab for a few years, but if you're new to the app and coding your strategies in C#, answers like this are definitely useful.

QUOTE:
Which UserStrategyBase methods are executed once per backtest?
The UserStrategyBase methods that are executed once per backtest are:
1. BacktestBegin: This method is called only once at the start of the backtest process.
2. Initialize: This method is called right after BacktestBegin and is executed once for every symbol being backtested.
3. Cleanup: Similar to Initialize, the Cleanup method is called for each symbol in the backtest, allowing you to perform any required housekeeping or logging on a per-symbol basis.
4. BacktestComplete: This method is called as a last step after the backtest is complete, allowing you to clean up any class-level static resources that were set up during BacktestBegin.

Using B126 w/Copilot.

Based on very early testing, it appears Q & A's are discrete rather than a conversation i.e. you can't ask a follow-up to its previous response. It would be useful to have a Conversation mode, w/o logging into one's AI account, that allows for pursuing the topic further. Concurrently, Clear/Refresh/New (or similar) option will also need to be provided so one can end a topic and start a new one.

That would for sure be a valuable enhancement. This is just a first step.

My summer intern develops RAG applications. I just saw this, but I was wondering if there is an easy way to download the WL8 documentation. I believe all the coded examples are easy enough to capture. In a short period of time , we have made good progress using Mistral and o4, Grok, Gemini Opus, etc. to do some complex tasks.

Is this something that can be done? It can be hosted on a website for others to use if WL8 finds it useful.

The WL8 help documentation is in the Help folder. Is there something more you're looking for?

Thanks. Let's see what these AI tools can produce.

QUOTE:
Is there something more you're looking for?

Well now that you're asking ... :)

Would it be possible to include "some" of the salient forum posts in the AI model?

Yes, I can think of a few to cherry pick from.

One idea to incorporate.

Allow it to search a community or users input or a personal directory on a person computer. The basic idea is as folks use this to search and get results, they can create a document or set of documents with the most relevant information and edit answers they produced. This becomes a feedback mechanism into the tool.

I've found this very useful when writing papers. For example, if I take a bunch of info, papers and patents and place them in a folder and say write a paper on xyz, then I get one output.

But if I then write an abstract based on the initial produced paper and incorporate that into the paper and resubmit it with a refined question, then it produces an even better result.

Continuing this process can result in a co-written paper using AI to act as an editor and information expert while the writer is more creator and producer.

Soon I will be trying this on my code base since my expertise is not in coding, so even I know it is in a bit of a mess.

There are endless applications when it comes to using AI as a tool.

Lately, I've been using ChatGPT and Grok with "code examples" formatted into ipynb/jsonl format for RAG to create code similar to some past examples I've created ("my style"). 2 questions:
1) Does/can the co-pilot have uploaded all of the sample code examples that are packaged into WL8 as part of its reference?
2) Feature request to add an "attachment" button so that we can also add our own RAG examples to further help with code generation.
This evolution of product will be amazing!!

Our co-pilot leverages the help files and blog posts, no code examples. A phase II AI integration would be to try and fine tune a code generation model with 100 or so examples of WL8 strategies so it could produce usable code from a prompt.

Do you plan to integrate Open AI as deep learning engine additional to TorchSharp and TorchSharp LSTM ?

No immediate plans for that.

Just started using the copilot help. It's quite slick!

I've tried code generation in ChatGPT and it's hit-and-miss. It does a good job for simple concepts but there are always errors to be fixed. One thing is I have to specify WL 8.0 - it defaults to 7.0 for some reason.