Mmm, not really unpredictable, it is easy for thought exercises to avoid the dirty real world, but in the real world, Large Language Models have recently proven that many quite difficult abstract probability exercises, when taken to the real world and applying training, can be succesfully solved well beyond our most wildly expectations of precision, speed and accuracy.
So yes, the prediction of the decision to have a ham sandwich tomorrow, for any given person is now, by the state of the art of applied mathematics and information science, a relative doable - if not plain easy - feat.
And according to some hypothesis, the brain could be doing this exactly kind of predictions, even more better than LLMs, with more accuracy/speed using way less energy.
> the prediction of the decision to have a ham sandwich tomorrow, for any given person is now, by the state of the art of applied mathematics and information science, a relative doable - if not plain easy - feat.
Sure, the LLM may be able to guess by reasoning (e.g. 'He always eats ham sandwiches for breakfast and he probably wouldn't break his routine'), the same way we could guess at another person's behaviour, but we're talking about reducing the brain into a deterministic input/output machine, which is a far larger ask.
Now if you said 'doable' that may be right in the long term, but 'easy'? Absolutely not. There's no current way to feed 'me' or 'you' into an equivalent LLM. The human brain has more axons then there are stars in the galaxy and we are nowhere close to even mapping these connections.
>The brain could be doing this exactly kind of predictions, even more better than LLMs, with more accuracy/speed using way less energy.
That's definitely an option. The fact that many of the brain's operations could be done by LLMs is a strike against the original thesis.
So yes, the prediction of the decision to have a ham sandwich tomorrow, for any given person is now, by the state of the art of applied mathematics and information science, a relative doable - if not plain easy - feat.
And according to some hypothesis, the brain could be doing this exactly kind of predictions, even more better than LLMs, with more accuracy/speed using way less energy.