3 ms·
I think Copilot has some hardcoded limitations of around a dozen files, like you said. But this stuff changes constantly.
by Paradigma11 1y ago
I think Copilot has some hardcoded limitations of around a dozen files, like you said. But this stuff changes constantly.
- ofrzeta 1y agoStill I don't totally understand how that huge of a context works for Gemini. I guess you don't provide the whole context for every request? So it keeps (but also updates) context for a specific session?
- Paradigma11 1y agoI dont know how the massive context works but Caching is certainly a thing and cheaper: https://ai.google.dev/gemini-api/docs/caching?lang=python https://ai.google.dev/gemini-api/docs/caching?lang=python Gemini is better than Sonnet if you have broad questions that concern a large codebase, the context size seems to help there. People also use subagents for specific purposes to keep each context size manageable, if possible. On a related note I think the agent metaphor is a bit harmful because it suggests state while the LLM is stateless.