Developing for the Amiga used to involve reading dense programming manuals and trial and error. In contrast, developing these days can be as simple as barking orders at ChatGPT to spit you out some Python code. However, that technique doesn’t work so well for Amiga languages, as ChatGPT hasn’t read much about the now-ancient platform. However, as covered by AmigaNews, there is now a ChatGPT model trained specifically on Amiga development. Enter Amiga Guru.
The work of [Cameron Armstrong], Amiga Guru was built after his early experiments with ChatGPT spat out non-functional gibberish when Amiga-compatible code was requested. The model has been trained on a corpus of official Amiga programming manuals, third-party books, and even the documentation for AmigaOS 3.2 and 4.1.
Using the model yourself requires a subscription to ChatGPT Plus, which prevents this writer from testing it directly. However, it makes sense that having been directly trained on Amiga manuals, it would be more capable at answering Amiga programming queries than conventional ChatGPT 4.
It’s easy to see the value of such a system. Learning to program for older platforms can be hard, with less resources available for new learners. Having an AI to help could be useful for some eager to develop for the 68K-based machine.
If you’d like to try Amiga Guru, you can access it via this link. Be sure to let us know how you go, and whether you think it has any value for speeding up your own Amiga development. Otherwise, if you’ve been doing anything else nifty with the platform that Commodore bought and paid for, don’t hesitate to let us know!
[Thanks to Stephen Waters for the tip!]
I haven’t tried using it for coding but it is giving me good advice on how to troubleshoot my Amiga 4000. Not a lot of places you can go to get that type of help.
This isn’t a hack and there is no explanation + examples how this works (the quality of training). So it is basically free advertisement? Boo!
(I’m not fun at parties or hackatons, I know.)
Yeah, I’d be more impressed if we had the data set and a low rank adapter for llama 2
Amiga developers must be very happy. Both of them.
Atari ST user detected. Hi there, li’ll fella :)
*looks up from her 386*
Was something?
Miss the good old days of the Amiga. CD32. All my respect to the developers who keep the legacy of this amazing platform alive.
Yeah… there are a surprising number of folks still working on Amiga, both software and new hardware. I’ve been to Amiga shows in Europe for the last 10 years or so and they get larger every year.
I feel like it’d be more interesting to see the *how* this was done… don’t get me wrong, it’s cool and all… but I suspect there are many of us that would like to similarly *hack* the LLM for our own particular purposes and more specific target scenarios. Maybe [Cameron Armstrong] could be encouraged to document the steps necessary to get to the results achieved. Additionally, it would seem to me that having such a trained model would be a *great start*, but knowing how to further *hack* this ourselves seems like a better tool so that it could be further fine-tuned over time (again, for our own specific needs).
My friend has a GPT subscription and he showed me how he trained GPT on extra data. It was really simple – you just log in, create a project and upload a bunch of PDF files containing the books you want it to train on. So I am guessing that the creator has taken all of the published Amiga books (which are available as PDF files on the Internet) and fed them into GPT. There might be extra steps that my friend didn’t show me, but that was basically it.
I have to agree. As cool as this is, it’d be a lot cooler if it were instructions on how to implement and run your own Amiga Guru on a RasPi or container.
Last week I picked up three A1200s, two of which have 68030 accelerators in them, from a late relative. Apparently my mum’s cousin has another two that I am welcome to! The relative (Graham) was in the local Cine-club and when I was 10 we filmed my primary school play and then edited it, titled it with Scala and a genlock on the Amiga, etc. then we all watched it at school. I’ll be 43 in a few days 😮
A little while ago I had been thinking about getting an Amiga (Nostalgia!) and all of a sudden I’m surrounded by them. When I was 13 or 14 we went to the Bowlers computer fair for my birthday and picked up an A1200 but when I got it home the keyboard was faulty and it was returned, much to my upset. I only ever had an A2000 (a kickstart 1.3, like an A500 in a big box) and then later an A600 with a hard drive.
One of my mates had a 500+ with an action replay cartridge. I’d love to have an AR on one of these A1200s with accelerator but it’s not possible.
I vaguely remember having some kind of resident software debugger type tool that you could interrupt the system with from a key combination, but my Googling powers have failed to turn anything up.
Anyway I need a PSU connector so that I can adapt an ATX PSU, I need a mouse converter, and I need a scan doubler before I can even turn them on, so it’ll be a little while yet.
I had been thinking about learning programming on them – and whether it is comoarable to MCU development (direct hardware, no mmu, direct framebuffer etc), but I doubt it. It would just be very tough I imagine. No fancy syntax highlighting intellisense linting IDEs..
Carl, absolutely you can have fancy syntax highlighting linting IDEs! Just not natively (although there are some that have syntax highlighting). Most people use something like VS Code on a PC/Mac/Linux, with a cross compiler, and they send the finished executable for execution on the actual Amiga. Sadly, you do miss out on the fun of actually using the Amiga to do the coding, but cross compilation does allow you to develop in a more modern way, which can greatly speed up development.