ChatGPT: ASCII to decimal conversion

[barnacle]

To be fair, I'd rather suffer bitrot on an eeprom than in my brain. That scares me, and projects like this keep my brain exercised. I wouldn't want it to be thought that I was considering giving up, just noting that there could be simpler ways to do it.

I started a formal German course this week; thirty hours a week for the next six months or so...

Neil

[BigEd]

good move - learning a language is said to be a good precaution against rot

[BigDumbDinosaur]

barnacle wrote:

I feel I could - with the knowledge I now have - probably write a C program to handle the major arcana in a week. Assembly, a little longer perhaps. C is, of course, assembly language with structures and 32-bit arithmetic.

Testing algorithms in C prior to reducing them to assembly language is a sensible route.  I used Thoroughbred BASIC to model my binary sequential time algorithms, which are very math-heavy, with 64-bit multiplication and division being used a lot.  Taking that approach likely saved me many hours of debugging, since when I finally wrote the assembly language equivalents, I knew my algorithms were correct—debugging was a matter of getting the LDAs and STAs correct.

Quote:

To be fair, I'd rather suffer bitrot on an eeprom than in my brain. That scares me, and projects like this keep my brain exercised. I wouldn't want it to be thought that I was considering giving up, just noting that there could be simpler ways to do it.

Although science hasn’t conclusively shown a direct link between lack of cerebral activity and dementia, I think there is enough anecdotal evidence to support the idea that the brain, in a sense, is like any other muscle: it will atrophy absent regular exercise.  Like you, I do things that require often-hard thought as a hedge against developing dementia.  Playing music, I think, is good for preventing cerebral bit rot, as is writing 65C02 machine code to beat on FAT32. 

Quote:

I started a formal German course this week; thirty hours a week for the next six months or so...

Wie Ed sagt, ist das Erlernen einer Sprache eine hervorragende Möglichkeit, die alten Synapsen zu trainieren. 

[Broti]

barnacle wrote:

I started a formal German course this week; thirty hours a week for the next six months or so...

Just remember: Deutsche Sprache schwere Sprache
But it's always good to learn new things. I started learning Dutch last May — 20 to 30 minutes every day.

P.S.: If you need some help, feel free to send a PM.

[cjs]

Martin_H wrote:

Go home AI you're drunk.

Yeah. My favourite one (which I think I posted here quite a while back) was when I asked ChatGPT for a simple "add two 16-bit numbers on a 6502" routine and it added the MSBs first, and then added the carry from that to the addition of the two LSBs.

While things are definitely improving, the LLMs are still nowhere near (and showing no indication of getting anywhere near) being anything more than an assistant to someone who already knows how to program and develop software. I pay for Claude and use it a fair amount, so as soon as I'd heard that the new Opus 4.6, with all the extended thinking etc. turned on, was finally going to do things right, I turned it on. I am still daily getting utterly stupid errors, as well as more subtle ones that I'm reasonably sure that someone without my level of experience won't be able to identify.

I guess, since it's been a couple of years, I should have Claude do another pass over some of my more highly optimised assembly code to see if it's any better, but I'm suspecting not. For basic syntax etc. you can expect it to do pretty well with extremely popular languages like JavaScript and Python, but I doubt it would do well with Algol, much less something so different as 6502 or similar assembly languages.

BigDumbDinosaur wrote:

I’m a bit amazed that AI would know enough about the 65C816 and its assembly language to produce anything of note, never mind whether it would work.  :D[

Assembly language is hardly unknown on the Internet. Now you might find a few 80386 instructions in your LLM-generated 65C816 code....

Quote:

I didn’t dissect the AI-generated code, but it seems to be a lot of code for what is supposed to get done.

Yes. LLMs tend to copy and paste code. A lot. Which is why I (amongst many others) currently have a job fixing the results of that. :-/

BigEd wrote:

We all know (surely) that any reasonable CPU can do anything that any other reasonable CPU can do, given memory and time.

Which is rather misleading, because of course what some CPUs have and others don't is precisely memory and time.

I mean, sure, you can use memory to substitute for registers in your 6502, and then after you've already increased your memory usage you can use yet more to deal with all the overhead of bank switching so you can get more memory, and so on, but even if you're willing to do al the work (including all the debugging), you've gone from a hundred times slower to tens of thousands of times slower than a VAX, and you're in that world where, theoretically, you can break any passphrase or encryption key because it's just a matter of time, right? (I think here is where that "sometimes quantity is a quality of its own" thing comes in.)

So, no, doing a thing that works fine on CPU X may not, in fact, be feasible on CPU Y, for any reasonable definition of "feasible."

[BigDumbDinosaur]

So it appears someone at WDC had been paying attention to all this chatter about AI writing code for us.  Predictably, they encountered a certain amount of nonsense of the type that has been discussed. 

[BigEd]

Note that the dateline there is June 2023 - whatever one thinks of LLMs, they are very different now, in March 2026.

[commodorejohn]

Dog Cow wrote:

That's because it can't invent anything new that it hasn't seen already, and it doesn't understand anything. All it does is remix text that it has come across on the Internet.

Still 100% factually accurate.

[Martin_H]

BigEd wrote:

Note that the dateline there is June 2023 - whatever one thinks of LLMs, they are very different now, in March 2026.

My recent attempt to use Claude to write 65816 assembly while better, still produced code that wouldn't assemble, couldn't link, and won't work. It's taken significant work slowly elevating the code to usable.

For the details of my misadventure see: viewtopic.php?f=9&t=8577