• minorninth@lemmy.world
    link
    fedilink
    arrow-up
    8
    ·
    1 year ago

    GPT-3.5 seems to have a problem of recency bias. With long enough input it can forget its prompt or be convinced by new arguments.

    GPT-4 is not immune though better.

    I’ve had some luck with a post-prompt. Put the user’s input, then follow up with a final sentence reminding the model of the prompt and desired output format.

    • vcmj@programming.dev
      link
      fedilink
      arrow-up
      2
      ·
      1 year ago

      Yes, that’s by design, the networks work on transcripts per input, it does genuinely get cut off eventually, usually it purges an entire older line when the tokens exceed a limit.

      • vcmj@programming.dev
        link
        fedilink
        arrow-up
        1
        ·
        1 year ago

        Or I should explain better: most training samples will be cut off at the top, so the network sort of learns to ignore it a bit.