• 0 Posts
  • 15 Comments
Joined 1 year ago
cake
Cake day: July 4th, 2023

help-circle





  • I have had some luck asking it follow-up questions to explain what each line does. LLMs are decent at that and might even discover bugs.

    You could also copy the conversation and paste it to another instance. It is much easier to critique than to come up with something, and this holds true for AI as well, so the other instance can give feedback like “I would have suggested x” or “be careful with commands like y”



  • I think I understand how it works.

    Remember that LLMs are glorified auto-complete. They just spit out the most likely word that follows the previous words (literally just like how your phone keyboard suggestions work, just with a lot more computation).

    They have a limit to how far back they can remember. For ChatGPT 3.5 I believe it’s 24,000 tokens.

    So it tries to follow instruction and spits out “poem poem poem” until all the data is just the word “poem”, then it doesn’t have enough memory to remember its instructions.

    “Poem poem poem” is useless data so it doesn’t have anything to go off of, so it just outputs words that go together.

    LLMs don’t record data in the same way a computer file is stored, but absent other information may “remember” that the most likely word to follow the previous word is something that it has seen before, i.e. its training data. It is somewhat surprising that it is not just junk. It seems to be real text (such as bible verses).

    If I am correct then I’m surprised OpenAI didn’t fix if. I would think they could make it so in the event the LLM is running out of memory it would keep the input and simply abort operation, or at least drop the beginning of its output.






  • I did back in college. Mobile computing was just becoming a thing but I was way too hipster (and poor) for a PDA or one of those newfangled “smart phone” devices.

    I hacked together a wifi SMS texting gadget following a tutorial on Hack a Day. It ran Debian with Linux kernel 2.6 and was so fun to tinker with.

    It had 32 MB of RAM but X used 11 MB of that so you couldn’t really do anything in graphical mode anyway. A shell running GNU screen however only took 4 MB so it was much more usable from the terminal.

    I eventually figured out a way to pipe images and even (non accelerated, since it didn’t have a GPU) video from mplayer to write directly into the framebuffer. It was a real bear to get it translated into landscape mode.

    I Am Legend in 144p never looked so good.

    Even with the terrible specs, I have never loved a phone so much as I loved that little computer