2 ms·
Not just languages. It's the case with everything. Easy to spot when you're tutoring someone. You can see they learned "force is mass times acceleration" or "fo
by TeMPOraL 3y ago
Not just languages. It's the case with everything. Easy to spot when you're tutoring someone. You can see they learned "force is mass times acceleration" or "for( ... ) is how you make the same code run multiple times" - they can tell you that when quizzed! But you know they haven't comprehended it until they can reverse it - "I need to compute the mass of the object, which I see accelerating this much under this force; I can pull that from F=ma -> m=F/a!", or "I want to run this code several times, I need a `for` loop!". And getting to that stage is often the longest and most difficult part.
- omoide 3y agoInteresting; your message hints at a sharp gap between using language stochastically (eg being able to repeat what you've heard) vs using language to express deeper knowledge/understanding. A common argument is that LLM by their nature only model the former, not the latter.
- TeMPOraL 3y agoMy point is that LLMs can do both, the same way humans can. We don't learn or recall reversals automatically either - we memorize them separately, or run an inference step (which, done often enough, leads to memorization). Reversals aren't free - not for LLMs, and not for our own minds.
- zarzavat 3y agoYes the difference is that LLMs learn statically: they train and that’s it. Humans dynamically: they revisit learned facts and synthesize new ones. If you force an LLM-like training process in a human, by doing rote memorization without any exercises, then you get LLM-like results in that human. This suggests that you may need to train an LLM on its own output. Goes against the conventional wisdom!
- angusturner 3y agoI think this “wisdom” is not really agreed upon, and in fact I’ve seen a few works successfully leverage synthetic data (including a cool one from Sergey Levine et al recently).