Opinions on Nigel?

[BowenM]

Updating to Q3 and I was prompted to install Nigel. I declined for now, but I was curious what those who have used it think? Is it worth spending the time and effort on?

Looking at a quick demo here:  https://www.youtube.com/watch?v=JF5q2Zjyx7Y it just looks like a bad version of ChatGPT. Amusingly, if you look his very first test case (2:00 mark) in the video showing how good Nigel is... the output is very wrong, but ChatGPT gets it right if you use the exact same prompt but tell it you're talking about NI products.

[Andrey_Dmitriev]

Unfortunately for me, Nigel is almost useless (sorry, NI).

What I expect from Nigel at least three major things:

• Context-Aware Code Suggestion — means Nigel should learn what my code is doing and then suggest the next changes, like I do manually. When I am programming in C, Rust, or Python in VSCode, GitHub Copilot just suggests the right next lines of code, so I hit Tab and Enter. Cursor AI also works pretty well. It looks like magic, but it works in many cases.

• Diagram Generation based on prompt means I should be able to ask something like "please insert binary search here," or "Create VI which turns a given JSON into an array of clusters," or "Add Timing check with HighRes Time stamp." This is what Copilot or almost any AI like Microsoft Copilot/Perplexity can do easily in almost any old-school text-oriented programming language, but not Nagel.

• Automated diagram errors corrections (both syntax and logic!) means if some wires are broken, then the AI should be able to correct and rewire it in the right way (not just trivial Ctrl+B). If I have errors or bugs in text code (caused by me myself or in AI-generated code), then Nigel should be able to fix everything, including logical errors (if I have a technically syntax correct Block Diagram but troubles in the algorithm).

Probably Nigel is good for absolute beginners who don't understand the basics and need additional explanation of the code, but for professional use, it is not enough.

[BowenM]

Thanks.

I kind of figured that would be the case. Does anyone have experience with how well the "analyze a VI" works? The demo claimed it was doing more than just analyzing the connector pane.

I also agree completely with your "three major things". Up until recently, LabVIEW was hands down my fastest language to program in. That is no longer the case because of the impact of AI.

[Xonmyth]

Hi BowenM,

I found it pretty useless. There is a limit to analyze up to 20 per month (at least that's the pop-up I got). It seems like it just describes whatever it finds on a VI BD.

Edit: Is it okay to share Nigel's response here?

I deleted it as im not sure if it is allowed or restricted.

[Xonmyth]

When asked "Is there something wrong in this VI?" It is mostly giving general advise. I think, it would work the best if the code is very well documented as Nigel will be able to read all the information and tell you what you have written. 

Edit: Same thing

[BowenM]

Yeah i saw the response before you edited. The results of analysis seemed maybe useful? But without being able to do something like track or modify state logic I'm not sure how useful it would be.

As for the "Is there something wrong with this VI?" query that seemed an entirely useless response. Well, uh...if you can't run the VI its probably busted. And if it does run did you make sure it does what you think it should?

[Xonmyth]

I personally did not find it useful. It just reads comments, captions, labels, and maybe something else in combination with general awareness of wires and structures to call out what's going on. 

It is "ok" for someone that has no experience with Labview and needs to inherit a code. 

When asked for how can I add 1+1, it gives you answer in plain text. 

And limit is something I saw for the first time and that makes no sense. If a newcomer wants to study a code, they will need more than 20 exchanges in a day. 

It may grow in future but not useful as of now.