On: The unreasonable effectiveness of an LLM agent loop with tool use

[voidspark]

I meant a verbatim exact copy of all documentation they have ever been trained on - which they are not. Neural networks are not designed for that. That's not how they encode information.

[Sharlin → voidspark]

That’s fair.