They’re confusing knowledge for function calling. If you say “what’s my nearest thingy?” It will call into a function in the app which will get your GPS coordinates to find a “thingy” near to there. and the result will be posted back to the LLM to serve you the result in human language. TheLLM doesn’t know anything that isn’t in the training / fine-tuning data, the context data, or function call results.
For those who believe AGI is right around the corner, this is just splitting hairs. The end result for AGI would be that it can discover your location easily, by making the function call, while still adamantly claiming it doesn’t “know”.
“I don’t know. I really don’t know, but THAT guy knows, and he’ll tell me if I ask. You didn’t ask if I can find out, only whether I know.”
They’re confusing knowledge for function calling. If you say “what’s my nearest thingy?” It will call into a function in the app which will get your GPS coordinates to find a “thingy” near to there. and the result will be posted back to the LLM to serve you the result in human language. TheLLM doesn’t know anything that isn’t in the training / fine-tuning data, the context data, or function call results.
For those who believe AGI is right around the corner, this is just splitting hairs. The end result for AGI would be that it can discover your location easily, by making the function call, while still adamantly claiming it doesn’t “know”.
“I don’t know. I really don’t know, but THAT guy knows, and he’ll tell me if I ask. You didn’t ask if I can find out, only whether I know.”