Anyone testing ChatGPT ability to rewrite JSON to create content?
I was going to test it, but with my limitations trying to avoid sharing content, it is kind of rough to test with.
I suspect that it could receive the json for multiple charts, and then reconstruct the data to significantly speed up my productivity.
Not sure what you mean with "rewrite JSON to create content" without an example but yea I've been using ChatGPT. I essentially treat it like a personalized stack overflow. It's pretty good.
Like if you feed ChatGPT a perspective page's json text of a page with some graphs that work in Ignition, I wonder if it can rewrite the tag paths and and graphs to show you the same data in another format maybe.
Because the json is effectively text, I thought maybe it would be possible.
I think I need to test it on my local sometime.
ChatGPT I believe was trained on stack overflow data (which I think was done without their permission and ethically dubious at best). I don't think it knows much of anything about Ignition at all.
I try to keep the language Ignition agnostic ie how do you do X in python 2.5. In your case you could just feed it the json, pretend you are only working with json and strings, and see what it can do then. Don't mention ignition or feed it data that is unrelated to what you need. Keep it as focused as possible for the best results imo.
I explicitly ask it in context of Ignition.
Helps me avoid things that I can't do in Ignition like functions in queries.
Yea I wouldn't do that. Just talk to it in terms of JSON. Can you modify this json to make these strings that are of format X into format Y.
I do what saves me time and is allowed.
I haven't tested the json yet. I think I will this week.
I think of it like a tool. I know someday, maybe it will do stuff without me.
I think for the next many years, it is just a tool that I need to get good at using.
It was trained using the GitHub repository, (without explicit permission from anybody with code there), and there is a lawsuit.
Edit: Here is the link to the article I read on this:
ChatGPT is a product of OpenAI
Would they not need to prove that they lost money as a result of these tools? Seems like an impossible task.
No. Copyright infringement has statuatory damages regardless. If a registered copyright (distinct from trademark), the statuatory damages are much higher. IANAL, yada, yada.
You need to be really sure of yourself if you copy something intending to claim "fair use". The standards are fuzzy, and seem to not be very generous.
I read this the other day:
ChatGPT is probably going to receive billions in investment dollars soon, so whatever the lawsuit yields, it looks like OpenAI will recover. I'd say that they are here to stay.
I am not particular about which AI can receive JSON and then give me back JSON.
It is a tool to me. I want to use a good tool to do work more efficiently.
I think for example a lot of my pages scroll right and left.
If it can convert them to scroll up and down fast, that would be ideal for me.
I am kind of happy to report that it couldn't write a query that pivots, has double joins, totals shifts for two different sets of people during each shift, which was about 1/4th of the total query complexity.
Ran across this today:
https://instapundit.com/565289/
So, when asking ChatGPT to write any code for you, and you don't really understand the topic yourself, how are you going to tell when it makes something up out of thin air? Possibly complete with false attribution to real experts?
I love this quote from that article:
Don’t fear the program that passes a Turing test. Fear the one smart enough to deliberately flunk it.
We always receive information from sources that can be wrong.
We always test the information, make a judgement, or do more research.
People use what they do know and the resources available to them.
I hope that when you were talking about someone not understanding a topic that you didn't mean me.
I am always learning more about SQL, Python, and Ignition.
If you do see some gaps in what I need to know, I hope you share with me the parts I need to learn.
If you mean people in general though, I think we test and seek resources to learn more.
I kind of treat it like a personalized stack overflow. "Show me an example of a decorator function that times the length of the inner function using python 2.5" will give you a nice example. Agreed you do need to know enough to proof read it and make sure it's right, but that's true of stack overflow or this forum as well, you should never just copy and paste code you don't understand.
Yes. Have you heard of "unknown unknowns"? Those are the things people don't know they don't know. From what I've seen from people deliberately poking ChatGPT on technical and scientific subjects, ChatGPT doesn't know much detail, but reports as if it does.
If the user doesn't know the topic, they likely do not know enough to test the result effectively.
I wouldn't trust ChatGPT as far as I could throw its datacenter.
Yah, good point.
In order to lower pressure, I often express:
"Even if I knew half of everything, I could be wrong in the other half."
I still didn't test the JSON capability.
ChatGPT is just another tool in the toolbox just like a Google Search or Stack Overflow has been for years.
Anyone using ChatGPT as a source of truth has already decided to use the tool incorrectly.
And if we want to take the example further then even doctors, engineers, programmers, professors get things wrong sometimes.
Very few things in the world are a perfect source of truth.