Recent

Author Topic: Testing Antropic  (Read 272 times)

Lauriet

  • New Member
  • *
  • Posts: 47
Testing Antropic
« on: July 18, 2026, 05:42:25 am »
Just In case anyone is interested:

I've attached a tool chain that I used Claude to develop. (Just to see if it could do it)
It is for an editor, compiler, simulator, HEXgenerator, code formatter for the 8052 microprocessor.
The compiler was real hard work for Claude....it kept getting things wrong, produced crappy assembler output. It took about 6 weeks of continual corrections to get to the stage where the output is reasonably efficient. If you didn't know exactly what the output should be, before you start, then the project will be a failure.
The simulator was likewise difficult for it, and it took MANY iterations to get it working. Once again, you need to know EXACTLY what the workings of it should be. If you just trusted it, then you will have very buggy software. (I wouldn't recommend NASA use it)
The other utilities where a bit easier. Probably because optimizations where less important, and the programs are shorter, or well understood.
So instead of developing software yourself, your job becomes one of checking/testing what AI produces.
As others have probably discovered, you need to be very precise in spelling out what you want, as a more vague request will produce more side affects.

Is it a good tool ?
It took 2 months, to develop there utilities and the question is how long would it have taken without it. Probably far longer, but the code seemed to be band aids on top of band aids, so maintaining it could be challenging.
The fact that it could do it at all was impressive, but it may need a few more years before its trust worthy.

If anyone wants to use them, good luck, and I take no responsibility for any remaining bugs.

440bx

  • Hero Member
  • *****
  • Posts: 6578
Re: Testing Antropic
« Reply #1 on: July 18, 2026, 06:42:28 am »
What you describe reflects my own experience with Claude.  You must know the subject extremely well because that is needed to detect and correct the many subtle errors it makes. 

Sometimes the errors are not subtle but, those are the easy ones since they tend to be fairly obvious.  The subtle ones, which are quite common, will be the troublesome ones for someone who only has limited knowledge of the subjects involved.

All that said, in spite of that, it is a great tool provided you know what you're doing.  It can definitely save an enormous amount of time.  Lastly, as you pointed out, being precise and exact goes a long ways to minimize problems.


FPC v3.2.2 and Lazarus v4.0rc3 on Windows 7 SP1 64bit.

Thaddy

  • Hero Member
  • *****
  • Posts: 19625
  • Glad to be alive.
Re: Testing Antropic
« Reply #2 on: July 18, 2026, 07:06:58 am »
All that said, in spite of that, it is a great tool provided you know what you're doing.  It can definitely save an enormous amount of time.  Lastly, as you pointed out, being precise and exact goes a long ways to minimize problems.
Yes. Be precise and the results are much better. You have to know the subject, but then Claude is quite good. For example, an UML modelled software design gets translated nearly 99% correct, but still needs careful attention. I used "Enterprise Architect" from Sparx to test that. Shameless plug: https://sparxsystems.com/products/ea/downloads.html

This is merely a software aid to make you more precise. I used it a lot in my former professional career.
« Last Edit: July 18, 2026, 07:14:19 am by Thaddy »
Any "programmer" that knows only one programming language is not a programmer

Tomxe

  • Full Member
  • ***
  • Posts: 146
Re: Testing Antropic
« Reply #3 on: July 18, 2026, 12:46:14 pm »
I can concur- Claude can help but you should use it to just speed up what you could do yourself.

pas8052 - nice, keep working!

 

TinyPortal © 2005-2018