• Replies 6 replies
  • Subscribers 18 subscribers
  • Views 392 views
  • Users 0 members are here
Options
Related

LLMs for Monkey C - What's Your Experience?

j.nygaard CPH DK
9 days ago

Hi folks,

I've been experimenting with using Large Language Models (LLMs like GPT, Gemini, etc.) for Monkey C, and I'm getting mixed results. 

I'm curious about the community's experience. Have you successfully used LLMs?? And which model seem to do better??

Specifically, I'm wondering about:

  • Generating reliable code (both whole/new project from scratch and adding to existing)

  • Debugging tricky issues

  • Upgrading to new SDK version. (e.g. the whole changing of variable typing from a while back)

  • Any prompting tricks you've discovered?

  • 8 days ago

    LLM-s are bad in languages that have few sources on the internet. Don't waste your time trying to convince an LLM that it doesn't know monkey c. It doesn't.

  • 6 days ago in reply to flocsy🤠

    Thanks for the insight.

    Yeah, these LLMs can seem great at frist, until you (most often) find yourself arguing with it about somethings thats obviously wrong. Like when I asked gemini about the standing of the Ryder cup a few weeks back. Kept giving me the result from hours ago and when I told it that was not correct, it then agreed and promptly gave me the result from last year. Out of curiosity and stubberness I kept at it for a while. It wasn't until I asked for the source I managed to get it to update...

    Somehow all the videos at youtube had me convinced they would be more useful as a dev assistant.

    I tried asking gemini about making an app that monitors sleep and vibrates if too much REM and it checked the API and came back and said stuff that seem correct. Like you can't see the sleep stages live and vibrating is easy. And when asked about using "stress" as an indicator for dreaming, it came back and said stress is based on HRV and might be a viable route....

    I get that they need at good code base to learn from, but I guess I was kinda hoping they could use their digital experience from C++, C#, java and others c-based languges, along with the official docs. Kinda like we would do, if we were new at the whole monkey business....  

  • 6 days ago in reply to j.nygaard CPH DK

    Look at the 2nd highest up-voted feature-request: https://forums.garmin.com/developer/connect-iq/i/bug-reports/feature-request-add-sleep-data-to-the-sdk

    Unfortunately many nuances are missing from the docs, that is only in common knowledge here in the forum. Go figure out anything from the forum...

  • 6 days ago in reply to j.nygaard CPH DK
    j.nygaard CPH DK said:
    monkey business

    LOL, I would guess that LLM-s know even less about "monkey business" aka monkey.jungle files Slight smile

  • 5 days ago in reply to flocsy🤠

    Thanks for the link. I hadn't really started proper research :-) 

    If I understand correctly, the stages are not computed until you wake up. Not sure if it's in the app or cloud. Either way, if that's correct, then such a SDK addition/feature wouldn't help much, as I need to affect the sleep real time...  

  • 5 days ago in reply to j.nygaard CPH DK

    Well maybe. Obviously this feature request was opened 3 years ago, so it's not likely it'll ever be implemented, and if it will then well, who knows how. Actually I don't know what sleep related data there could be in the SDK, so I couldn't ask more precisely, but my intention was (maybe it's not clear from the ticket) to be able to make apps that can get the data in real time and decide when is the best time to wake you up. Of course there can be many other things in the way, like I'm not sure a background app would be able to trigger a wake-up alarm of some kind (probably not like when you set an alarm on the watch, and currently my guess is that not even sound and vibration)