• Buddahriffic@lemmy.world
      link
      fedilink
      English
      arrow-up
      1
      ·
      1 day ago

      There is feedback that makes it back to their servers: the next prompt.

      Though from my understanding, a lot of the agentic tools do a lot of prompt generation, so that would be a weakness of just logging everything and using subsequent prompts to evaluate earlier ones. They’d have no idea whether each prompt is coming from an actual user or another LLM.