sbp commented on issue #14: URL: https://github.com/apache/tooling-agents/issues/14#issuecomment-4352850338
> the agent can call its invokable LLM N times in sequence Yeah, but the main aim is to use cheaper LLMs to gather the context, medium cost LLMs to collate and do a small amount of reasoning about that gathered context (also compactifying), and then high cost LLMs to work on the output of that. This is covered in the second section anyway, so I think the first section is just irrelevant. > if good return it Nope, the early phases are for gathering. It's not like we're trying weaker models on the same task and then trying it over and over with more expensive models until there's success. The reason why we can't do that is because verification may cost nearly as much as production in the first place, and how can you get a weak model to know whether it's done the right thing or not? > This one's a math technique Well, [yes](https://en.wikipedia.org/wiki/Telescoping_series), but the missing section here is about compactifying inputs but then caching them outside of the tools and agents. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
