>retards ITT expect models to "know" things and have accurate information
You're all literally braindead. You only need a model that has solid "reasoning" and decision-making skills to leverage tools like web search or whatever is appropriate for the task at hand. I couldn't give a fuck less if an LLM gets the question "What is the capital of Wisconsin?" wrong as long as it's capable of interpreting/executing on my instructions and then working with the responses it receives from tools correctly. The transformer architecture is not a database or a wiki - LLMs don't "know" anything and wanting them to without hooking them up to reliable tooling is dumb as hell.