I wrote support regarding this concerning situation on the forum and I received a good and prompt answer from support that may be good. Kimberly from support provided this email scriptingsupport@ninjatrader.com which I definitively like more than a general support@ninjatrader.com. It remains unclear why the search is improved but now we won’t be able to search through the answers that they’ll give to their customers and I’ll ask.
But unless the user takes the email response and posts the issue and the solution every single answer is done on a one on one basis… so no searchable database of answers, problems or solutions. Just an enormous backward step.
And can you imagine answering the same issue over and over and over…
3 Likes
See here for AI response from Support:
It does say ‘copied to Support’ so we’ll see if that brings any ‘real’ intelligence
For completeness - no reply from Support
This AI support is terrible, only going to see more and more companies going this way sadly.
yes i use chatGPT to code but 10% of the time it is just wrong and another 10% of the time it will take a scenic route - in one instance taking 67 lines of code to achieve what could be achieved by me in 1!!
And it always says yes that’s the right answer which can get infuriating - but that’s useful to me - it shows me the time to step away and smell the grass!
1 Like
The trick is, you have to become familiar with what it gets wrong. When you know this it’s easy to correct. But yea but does lie all the time telling you the code is working yet it has no way of checking it. Also it regularly makes up NT functions that don’t exist.
It’s almost perfect with basic C# but when you start Using OnRender and ChartControl it gets worse.
I think you have to go in not expecting it to code a whole indicator for you. But you have to work With it and go back and forth.
I use both Claude and ChatGPT. If one gives an answer I’m unsure of I’ll ask the other.
1 Like
yes it absolutely loves SharpDX.Mathematics! - must have told it 10x you can’t in NT but it keeps insisting. Guess it will update eventually.
Here’s what might help.
Feed the ninjascript documentation to ChatGPT, give it the links, and it will learn.
I’ve built out institutional grade systems for my retail setup, first by thinking up the idea, discuss it with AI to give it context, and then feeding it what it needs.
ChatGPT will still struggle with that though because a lot of the NT8 documentation has been rewritten and is located at new addresses. If GPT doesn’t know to search there, it might never find it. A better approach would be to find legit code from github or old examples from the NT8 support team and compile them into one big document. Then just send it to GPT as “example references”.