I am looking to automate the creation of module chains with LLMs. If I feed a scenario blueprint to a capable, large context window model like Claude 3, it can read and understand the blueprint. It could also, in theory, print other modules according to my specifications.
Unfortunately, many modules create huge blueprints, which is too much for the output tokens of even Claude Opus or Gemini Pro, so I would like to reduce the blueprint’s footprint, if possible.
- If I were to systematically reduce the amout of info in a module’s blueprint (like default or empty key value pairs), change my variables, then re-import to MAKE, would it still work?
- If so, which parts could be removed without breaking the whole thing?