That would require function and intelligence far outside the bounds of current large language models.
These are models. By definition they can't do anything. They can just regurgitate the best sounding series of tokens. They're brilliant at that and LLMs will be a part of intelligence, but it's not anywhere near intelligent on its own. It's like attributing intelligence to a hand.
Except it’s already been shown LLMs can do exactly that. You can prime the model to insert something like ${API CALL HERE} into its output. Then it’s just a matter of finding that string and calling the api.
Toolformer does something really neat where they make the API call during training and compare next word probability of the API result with the generated result. This allows the model learn when to make API calls in a self supervised way.
The model can be trained to output tokens that can intercepted by the backend before returning to the user. Also, the model can take metadata inputs that the user never sees.
Yes. It is possible to do additional things with the model outputs or have additional prompt inputs... That is irrelevant to the fact that the intelligence -- the "trained" part -- is a fixed model. The way in which inputs and outputs are additionally processed and monitored would have completely different intelligence characteristics to the original model. They are, by definition of inputs and outputs, separate.
Models of models and interacting models is a fascinating research topic, but it is nowhere near as capable as LLMs are at generating plausible token sequences.
These are models. By definition they can't do anything. They can just regurgitate the best sounding series of tokens. They're brilliant at that and LLMs will be a part of intelligence, but it's not anywhere near intelligent on its own. It's like attributing intelligence to a hand.