coolboydev

I wouldn’t use an LLM as a one-off chat. I’d turn the research into a long-running Codex or Claude Code project, keeping the facts, findings, decisions, and actual outcomes as project documents. There usually isn’t an “optimal” answer from day one. But as the project builds a history of what you believed versus what actually happened, the LLM can learn from it, refine its judgment, and gradually get closer to the optimal solution.

This shows just how important open-source language models are.