Two Stop Bitsnew | comments | tags | ask | submitlogin
I tried vibe coding in BASIC and it didn't go well (atari 8bit basic 800) (goto10retro.com | ia)
11 points by goto10 50 days ago | 9 comments
  • mmphosis 49 days ago
    As an experiment, I just tried what you did using Applesoft BASIC and it didn't go well either.

    https://mmphosis.netlify.app/10print/llm/

    • mmphosis 48 days ago
      I tried Vibe coding in BASIC using Grok and it went well.

      https://mmphosis.netlify.app/10print/llm/grok/

    • masswerk 49 days ago
      This is much more like I would have expected this. – Thanks for risking your sanity in the name of sicence!
      • bmonkey325 49 days ago
        Oh man. that was an awesome read. I am tempted to post your blog post as a top level post :-)
        • masswerk 49 days ago
          The honor is all @mmphosis', though. :-)

          But I second the idea.

  • bmonkey325 49 days ago
    @mmphosis @masswork - not really surprised that the LLM doesnt know Applesoft or Atari BASIC. Even if all the print material produced in the pre-internet era was machine readable, it would be dwarfed by the amount of visual basic and VB.NET material that was put on the internet in blog posts, tutorials github. Even the old print material wouldn't always be precise. "Here is this months game - written in BASIC"....only because it as InCider or Analog or Antic would you know by meta inference that it was the specific BASIC.

    As a preservationist, it now gives me a motive to start putting retro code up on github to reinforce the nitty details of what we lost in the fire with Commodore, Apple II and Atari 8-bit machines....

    • masswerk 49 days ago
      On the other hand, there are probably about as many references to BASIC not being just BASIC, but rather referring to a variety of dialects, as there are digitized listings. Moreover, about every of those listings is accompanied by a reference to the specific dialect it applies to. If there was a world model, it should be clear that these are not the same identical reference and those rather refer to a complex set of is-a / has-a relations. The actual outcome illustrates a serious limitation of LLMs with regard to real-world tasks.

      (It also seems to hint at a general issue with intersecting partitions and, I guess, the same would apply to JS in the era of the "browser wars", say, something that runs on both Netscape 4.0x and the various incarnations of IE4.x, to make it easy.)

  • bmonkey325 50 days ago
    Cool idea. Judging by the title, I was expecting abject failure. Maybe I am forgiving in my expectation of current state of the art of LLMs. I have no idea what the context prompt was for this. Was was as elaborate "act as an atari 800 programmer, draw on sources such as De Rey atari by Chris crawfor et al and all of tom hudson's game and utility programming listings in analog computing for the years 1981-1989 - using this knowledge please help me write a game that has the following...."

    I mean the game shown was far better than some of the early games I wrote in the late 70s on a TRS 80 at computer camp or on my own Atari 800 in the early 1980s during the lengthy Canadian winters.

    Maybe I am soft with nostalgia ...

    • masswerk 50 days ago
      While it worked out better than I would have thought, it also points out some crucial problems.

      The problem being, if you don't know the answers already, the LLM doesn't fill any gaps in knowledge, you may have. E.g, I don't know Atari BASIC (just that pesky – or is it petcii? – MS BASIC). Accordingly, I have no idea about PRINT #6 or LOCATE, and ChatGPT apparently neither, so the duo of us won't arrive anywhere.

      On a slight tangent, I'm kind of surprised and impressed to learn that Atari BASIC lets you use a keyword as a name with LET, something MS BASIC won't do, tokenizing it anyways, LET or not, as there is no such thing as context awareness.

lists | rss | source
Search:
Two Stop Bits is a discussion web site about retro computing and gaming.