I think what he's saying is that an accuracy of, say, 10 mg/dL on a continuous basis is better than an effortful (requires you to test — and, perhaps more importantly for Type 1 diabetics, to be able to test) accuracy of 1 mg/dL. That's still going to tell you with enough accuracy whether you need to get some carbs into your system, or whether you've put too many in already.
I think that would depend on the type of errors the device makes. If every 1s measurement is of by independent errors, you can simply do a running average and improve accuracy that way. If the error is fixed, you can determine it and compensate.
If, on he other hand, errors between close in time measurements are correlated, that won't work. You might get "things look fine" measurements for an hour while your glucose level is dangerously low or vice versa.
Given that the first two kinds of errors can be worked around, I would guess this device to produce the third kind of error. You don't need 1s intervals in glucose reading to improve the life of diabetics; 10 minutes or even an hour would be fine, too.
I guess most meters have a display accurracy of 1 mg/dL, but that's just on display. If you check your blood twice in a row, you're likely to get at least a 5 mg/dL difference. I'm not even sure this is your meter's problem, but rather your blood having not 0.01 mg/dL-consistent sugar across all of your body.
Nevertheless 10 mg/dL is just the right accuracy needed to control type I diabetes. I don't tend to attach much importance to the last digit anyway.
Reduced accuracy would be offset by frequency of feedback. The current model requires active input; you have to go and prick your finger. With a passive input, you could get a general idea all the time, which is especially important considering that a low blood sugar can sometimes affect your mental capacity.
Please correct me if I've misconstrued.