Teach a man to fish..... (ChatGBT / AI)

[ILoveMadoka]

"Give a man a fish you feed him for a day...

Teach a man to fish and he can feed himself for the rest of his life.."

 

I am a novice (if that) at writing code.

 

I just found that CHATGBT (ai) can write autolisp code. ChatGBT

You have to be able to verbalize what you are after using decent descriptions (which mods have to dig for when helping end users such as myself define our needs)

I'll admit it misses the mark more than not but what I have found <for me> is that it gives me enough of a foundation that I can troubleshoot it myself and get a routine working...

 

Wanted to let everyone know there is a tool that can help those of us who are "learning to walk" as it were..

Even if it's a simple as how do I do "X" in lisp...

I'm hoping that like other AI sites that it will get better in time...

 

If you know nothing, probably not going to help since you cannot troubleshoot...

If you know a lot you'll see the coding errors right off and think it's a waste of time...

If you know a little and are trying to learn it is a cool resource...

 

I used it to write some Powershell code and it did a really good job.

 

FWIW..

[mhupp]

1 hour ago, ILoveMadoka said:

If you know a lot you'll see the coding errors right off and think it's a waste of time...

If you know a little and are trying to learn it is a cool resource...

 

I wouldn't say a wast of time. and its only like what 2 or 3 months old? just imagine what it will be able to do in like a year or two.

 

cool resource indeed but if using it to learn coding. keep in mind (just like here) it might not be giving you the right information. but the chat isn't going to correct itself. Were as if someone post here in error or if someone doesn't agree with the method its a good chance to get a follow up post.

 

Quote

 

"The best way to get the right answer on the internet is not to ask a question; it's to post the wrong answer."

Cunningham's Law

 

 

 

 

[ILoveMadoka]

Wrong more than right does have it's consequences...

Banned on StackOverflow

[j2lstaples]

Its recent version drastically improved in terms of its code suggestions, and it became better at coding in LISP than what it was a few weeks ago. I know because I've been asking it basic LISP questions and correcting them and found that it had less errors now than before. As it becomes better and better at logic-based problems, the syntax in the solution will come next. It's an incredible resource just to get started with. Hopefully, it gets better and be implemented in multitude of tools like IDEs and even Visual Studio. I think even recent versions of Windows already implement its autocomplete function. I imagine that for eternal students of code and software development, it would be a common tool.

[BIGAL]

Interesting given it costs to use so if any of us say more experienced get asked here to fix code does that mean we will get paid also ?

Edited February 22, 2023 by BIGAL

[Steven P]

Here is a thought about AI, it still needs a lot of checks and verification that it does what it says it does. It is possible for it to create a LISP that appears to change a design document as desired and also change things unexpectedly.

 

So who is responsible when "the computer told me to do it".. If I have been paid professionally to make a LISP and it goes wrong, then it is my fault. If an open source thing makes a LISP then responsibility is no one?

[Abrasive]

My opinion is that it's a great tool for a knowledgeable person but could be problematic for a novice like me.

Because of my limited knowledge I would run the risk of having bad code, or worse, harmful code.

BUT... in the right hands it's a great advancement.

[Steven P]

I'd agree with that.

 

A little knowledge can be a dangerous thing, for example you know you want a LISP but not enough knowledge to fully test an AI generated LISP (or even know you should).

Then you get enough knowledge to know you don't know stuff - known unknowns you might call it

Then you get more knowledge and know - or can find out - all you need to know - known knowns

 

But it is the one at the beginning of the learning, who has unknown unknown stuff to deal who could be making mistakes deciding that the computer knows everything.

 

 

 

Am sure there is quote with that in somewhere - well before my time obviously.

[BIGAL]

Code working on at moment is 1100 lines and getting bigger, would be interested in how AI would work in that case writing the  defuns with lots of if's and buts required.

Edited February 25, 2023 by BIGAL

[Steven P]

Try it and see!

[BIGAL]

What about DCL's ? If it could do those then would be a big advantage. Just not sure need to see say a complicated task duplicated so can see existing code and what ChatGP does and is it worthwhile to use.

 

What about this done in ChatGP make a rectang and dim it.

 

; simple draw a box and dimension it 
; By Alan H March 2019


(defun ah:box ( / pt1 pt2 pt3 ahl ahh ahoff )
(setq oldsnap (getvar 'osmode))
(setq oldang (getvar 'angdir))
(setq pt1 (getpoint "\nPick lower left"))
(setvar 'osmode 0)
(setq ahL (getreal "\nEnter length "))
(setq ahH (getreal "\nEnter height "))
(setq pt2 (polar pt1 0.0 ahl))
(setq pt3 (polar pt2 (/ pi 2.0) ahH))
(command "rectang" pt1 pt3)
(setq ahoff (* 2.0 (* (getvar 'dimasz)(getvar 'dimscale)))) ; change offset as required
(setq pt4 (polar pt2  (* pi 1.5) ahoff)) 
(command "dim" "hor" pt1 pt2 pt4 "" "exit")
(setq pt4 (polar pt3 0.0 ahoff))
(command "dim" "Ver" pt2 pt3 pt4 "" "exit")
(setvar 'osmode oldsnap)
)
(ah:box)

 

 

 

 

Edited February 26, 2023 by BIGAL

[j2lstaples]

I even started using Bing Chat which uses GPT4. So far it gets the syntax right, however the functionality might not be what you want. There are ways to coerce it to read its own generated code and do basic debugging. It's extremely useful for asking questions like, what does this code do. As you progress further in terms of using LISP, you see a lot of the errors but in order for it to be better, you have to respond to it after you test its code so that it becomes more logical. I also ask it to reference websites like Lee-Mac's when I have an issue with one of his codes. This is a process called RLHF (Reinforcement Learning with Human Feedback). In order for it to be truly useful for LISP, you must test it and confirm if its code works. Due to lacking training data of LISP, it was very bad at the beginning.

[BIGAL]

As I have said before can not remember which CAD it was but its Macro record wrote VBA code, may have to fire up my 10 year old lap and have another look for it. This was real useful code. You could edit it and huge resources for VBA.