3 ms·

You could develop an AI model to help pick the correct AI model. Now you’ve got 18 problems.

• skygazer 2y ago
  I think you're trying to re-contextualize the old Standards joke, but I actually think you're right -- if a front end model could dispatch as appropriate to the best backend model for a given prompt, and turn everything into a high level sort of mixture of models, I think that would be great, and a great simplifying step. Then they can specialize and optimize all they want, CPU goes down, responses get better and we only see one interface.
  • louthy 2y ago
    > I think you're trying to re-contextualize the old Standards joke Regex joke [1], but the standards joke will do just fine also :) [1] Some people, when confronted with a problem, think “I know, I’ll use regular expressions.” Now they have two problems.
  • calmoo 2y ago
    Isn't this basically the idea of agents?
    • skygazer 2y ago
      I don't believe so. I thought agents were go-do-that-complicated-interactive-thing autonomously on my behalf. But, more similar to tool use, except, with mixture of experts, each expert assumes the continuation of "participant identity" in the conversation, in that they're fed the whole context.
      • calmoo 2y ago
        Yeah you're right, I had a misunderstanding of the term.