RPN programming with Open AI Gpt

[Roberto Volpi]

Hi all

testing the capacity of Open AI Gpt I asked it to make an RPN program to compute the determinant of a 4x4 matrix, and I received the answer in the picture.

Maybe it cannot compile appropriate coding, but can help find a quick and dirty algorithm.

Do you think it can be of any real help?

[Valentin Albillo]

(01-18-2023 02:20 PM)Roberto Volpi Wrote:  testing the capacity of Open AI Gpt I asked it to make an RPN program to compute the determinant of a 4x4 matrix, and I received the answer in the picture.

Maybe it cannot compile appropriate coding, but can help find a quick and dirty algorithm.

Do you think it can be of any real help?

As of January 2023, no, for non-trivial math code it's of no help at all, more of a hindrance as you'll spend more time and effort debugging it to make it work properly than if you created proper working code yourself in the first place.

In the (near) future, who knows ...

V.

[Thomas Klemm]

Related: Write a GCD program for the HP-41C with chatGPT

[floppy]

A meta interpreter for RPN could be done in PROLOG (on my ToDo list. Done when? dont know..)
https://www.metalevel.at/
https://www.youtube.com/watch?v=nmBkU-l1...erofProlog
Lisp interpreter exists https://www.metalevel.at/lisprolog/
And for another example https://www.metalevel.at/tist/ see interpreter https://www.metalevel.at/tist/interp.pl

[floppy]

(01-18-2023 05:40 PM)Thomas Klemm Wrote:  Related: Write a GCD program for the HP-41C with chatGPT

Hilarous. For me looks like somebody in this forum could contact chatGPT and offer them to train their KI on HP41.
Then behave like an Apple employee: dont go below 300.000USD per year (and more: ask for maximum 40h per week. HP41 is a technology edge topic like recovering dinosaurs bones). haha.
Never ever make such a job (for a foreign company) for free.
This is my view and my personal recommendation.

Would be interested to know how much money Thomas Klemm received from Microsoft for having trained its KI on HP41 for few hours (haha).

[Maximilian Hohmann]

Hello!

(01-19-2023 09:16 AM)floppy Wrote:  Never ever make such a job (for a foreign company) for free. This is my view and my personal recommendation.

Yes and no. Most will know that the main shareholders of OpenAI are Elon Musk and Microsoft/Bill Gates, so two of the richest persons of this planet (the first one not any more so much since a few weeks). OpenAI is certainly profit oriented, even with the "open" in it's name. But there lies the problem: From a steadily decreasing handful of RPN enthusiasts there is not much profit to be expected. HP has not given up it's calculator branch without a reason... Which means that OpenAI is not going to invest a single cent into training it's networks with RPN calculators in mind. They rather do without this tiny little share of the market.
So if we want Open AI to be able to spit out RPN programs it is going to be us to do the work (for free) or otherwise it will never happen. Everybody has to chose for himself of course!

Regards
Max

NB: At first glance, without going into details, the matrix inversion program does not look totally wrong even if it will certainly not work. Did anybody actually check what results it produces?

[Roberto Volpi]

(01-18-2023 05:40 PM)Thomas Klemm Wrote:  Related: Write a GCD program for the HP-41C with chatGPT

If this Open AI is of any use for our scope, it is just for some hint in the general algorithm

Negotiating amendments with an AI is not useful. I dare say a complete waist of time, unless we wish to have a good time with it.

[Valentin Albillo]

(01-19-2023 03:32 PM)Maximilian Hohmann Wrote:  NB: At first glance, without going into details, the matrix inversion program does not look totally wrong even if it will certainly not work. Did anybody actually check what results it produces?

What "matrix inversion" program are you talking about !?

It's supposed to be a "determinant" program but actually is but pure, unadulterated garbage.

I don't get how people expect a glorified nincompoop system, which regurgitates text from a condensed megarepository, to be able to create a bugless, working program (when we know that any simple typo or omission will cause a non-working piece of code,) so the mindless regurgitations of this "A.I." is sure to require more effort to debug than to create a proper program anew.

Unless the piece of code you are requesting already exists in the repository, you'll never get anything but a steaming pile of manure, mainly because this thing doesn't think nor does it understand anything, it's nothing but a multillion-dollar parrot, as far as creating new programs is concerned.

Regards.
V.

[Maximilian Hohmann]

Hello!

(01-20-2023 05:33 AM)Valentin Albillo Wrote:  What "matrix inversion" program are you talking about !?

Pardon me, I was thinking of one thing and have written the other... matrices have not been part of my life any more since at least two decades.

Calculating a determinant makes it even easier for the AI, because this can be expanded into an equation, either by looking it up in a mathematics reference book (a digital one of course) or by getting it computed from a CAS algorithm.

All what needs to be done thereafter is to reformat the resulting equation into postifx notation comaptible with the programming paradigm of an HP35S. This is a rather mechanical task that requires no artificial intelligence at all.

(01-20-2023 05:33 AM)Valentin Albillo Wrote:  I don't get how people expect a glorified nincompoop system, which regurgitates text from a condensed megarepository, to be able to create a bugless, working program (when we know that any simple typo or omission will cause a non-working piece of code,) so the mindless regurgitations of this "A.I." is sure to require more effort to debug than to create a proper program anew.

I am much less skeptical in that regard. What we see here is of course still very incomplete. But whenever I drive a car in difficult conditions (darkness, snowfall, slush on the road, barely visible markings, heavy traffic, cyclists and pedestrians in dark clothes on and close to the road, ...) and get the feeling that it would be best to park the car now and continue my journey by train, I think about autonomously driving cars that can easily handle all that and much more.

I have the feeling that such an autonomous driving system is much more complicated and requires much more programmed "intelligence" than a system that transforms a mathematical problem into a sequence of RPN keystrokes.

Regards
Max

NB: This is the expanded form of a 4x4 determinant. To me, without having checked it term by term, it does not look very different from the AI-result in the first post.

[Thomas Klemm]

(01-19-2023 09:16 AM)floppy Wrote:  Would be interested to know how much money Thomas Klemm received from Microsoft for having trained its KI on HP41 for few hours (haha).

About the same I got spending 398.4 traditional American workdays ^[0] on this forum.
I ended up having more success than training another AI bot recently.

We've known for a while that: "If the service is free, the customer is the product."

---

[0]: I didn't bother to figure out how much time I spent on the old forum.