

That’s probably the worst possible addition. Something like this, you need to be able to depend on. In a no-room-for-errors kinda situation you really don’t want to have a language model hallucinate something and burn potentially ruinous amount of scarce resources in the process, not the least being time. For example with crops.
Edit: and also that’ll burn through the power compared to just reading pdfs… if that’s scarce too, it’s a no go. Not to mention it’d have to really have some bulk for the capability to even run a basic model with extremely hit and miss results and definitely zero chance of retaining any sort of context long enough to be actually useful in this kind of use case. I think people are probably a little bit pampered by the cloud models of today. No chance you’ll be running anything like that in a small device with limited power.
Well, I guess sufficient is subjective…
However, consider that whatever you have now in today’s world, not just physical things but also your life experience and expectation of things might not line so well with those that you’d have in a scarcity scenario completely unlike today’s world. What you now consider sufficient might be entirely unacceptable in different frameworks of thinking and living.