After a recent contribution by fellow columnist Eef, suddenly a lot of articles in the Dutch and Swedish press about this AI were published – did they read Eef’s column? As usual with the presentation of any new AI, the articles vary from handy tools for the curious, via a new don’t-have-to-do-any-more-homework trick for students, to the latest threat for AI to take over our jobs and ultimately the world.
Finally, I got curious and decided I had to try the thang for myself.
Of course, you cannot just log on and start asking questions, that has already been possible by the Hi Google and similar AI’s. No, as clearly shown by Eef’s effort, this thing is something else, and any testing should be done by creating some sort of dialogue.
To get started, you need to surf to the AI here, and then look for the ‘Try ChatGPT’ button. Press the button, log on or create a profile and then log on, and you’re in business. Being new to this AI, I had to create a profile.
While doing so, you have to answer a couple of questions, where one, in particular, struck me as odd: I had to declare that I was not a robot. That’s perhaps the AI variety of one AI telling the other to mind its own business. Anyway, everything seemed to go smoothly and within a minute I was up and running.
Now about what to ask the thing… Where Eef already covered the philosophical side of talking to the AI, I decided for a more practical approach.
Since 1990, I have been earning my living as a programmer on a specific platform made by IBM, using a programming language specific to this platform. This language, called RPG, is about 50 years old and has transformed from its archaic punch-card format, via fixed format to its current free format. Needless to say, I use the newest Free format variety. With the IBM platform perhaps a tad obscure, and the RPG language even more so, this topic would be the first to interrogate the AI on. Would it know the platform and the language? If so, could it help me write me a program?
As I’ve written and published a couple of books, I am curious if the AI would know of them, and if so, how much. This might make a good subject for the next conversation.
With our holiday home in Los Alcazares, Spain, it might be nice to find out more about the place by asking the AI for some hidden gems out there.
Let’s get started!
- Here you can read about the RPG conversation.
- Here’s a short one on the AI reviewing my books.
- Here I asked the AI about our summer residence Los Alcazares.
- Things can go awry!
Once more an impressing performance of our by now well-known AI.
Let’s give ‘it’ a name; what about Art-I or Arty?
And because of Arty, ‘it’ is a him.
What can we learn from Paul ’s questions and from Arty’s answers?
To the request ‘how to write an CL program that reads a file ‘, Arty delivers a straight forward example with a GOTO leap.
Paul guides him into a better structured ‘modern CL without using GOTO’ and Arty provides without the briefest pause an example with a DO-loop.
After that Paul suggests a FOR-loop and there it is.
When he interrogates Arty about the preference he just showed – from GOTO to DO-loop to FOR-loop, the AI apologizes about this with an elaborate story on the ‘personal reference of the developer’.
It seems obvious that Arty doesn’t optimize his answers upfront.
He ‘reads in’ as it goes. Taken the first example he encounters, probably from a tutorial for beginners and he does not evaluate the practical application of this code.
When asked – and guided – into a better approach, he reads further and encounters more sophisticated methods and brings these out.
This is an extremely important conclusion.
Paul has demonstrated with this dialogue that Arty is NOT the know-it-all that lots of students think he is.
His answers are based on what he finds in the knowledge base that he has access to, but his response is always related to the question.
Arty doesn’t have this knowledge embedded in him, he just is incredibly fast in finding stuff.
But again: the question determines the integrity of the response.
So when the question is limited, the answer will be.
For obtaining a fully adequate RPG-program, the assignment will be so extensive that:
- It requires a skilled programmer to compose such a question,
- Being a skilled programmer, it is probably faster to write it yourself.
Eef 4 feb 2023