3 ms·
It has to learn the meaning of words, including implicit associations, and to do that it needs to see approximately all the English text ever. We don't know how
by astrange 3y ago
It has to learn the meaning of words, including implicit associations, and to do that it needs to see approximately all the English text ever. We don't know how to balance this with only feeding it useful knowledge.
- eru 3y agoIt doesn't necessarily have to see approximately all the English text ever. Real people don't learn English like that, for example. It's just that given what we know about neural networks, it's often easier and simpler and more effective to increase the amount of training data than to change anything else.
- basch 3y agoReal people might not speak every dialect of English. They may not be well versed in local grammatical oddities.
- marshray 3y agoDoesn't seem to be much of a problem in practice? If someone knew all the math and science in Wikipedia, for example, I think they'd probably be forgiven for not knowing every regionalism.
- astrange 3y agoUnfortunately models aren't always good at knowing what they don't know ("out of distribution data") so it could lead to confidently wrong answers if you leave something out. And if you want it to be superhuman then you're by definition not capable of knowing what's important, I guess.
- eru 3y agoBtw, models like GPT 4 can express that they are not confident. But that looks like a 'smeared' out probability distribution on the next token. Not like the text produced by an unsure human.
- user_named 3y agoAn LLM has nothing in common with a human. An LLM works in ways that have nothing in common with the human brain.
- eru 3y agoYes, LLMs and human brains share at most some faint similarities. Nevertheless, human feats can act as an existence proof of what is possible. Including of what might be possible for a neural network. (I'm not sure whether a large language model necessarily needs to be a neural network in the sense of a bunch of linear transformations interleaved with some simple non-linear activation functions. But for the sake of strengthening your argument, let's assume that we are assuming this restrictive definition of LLM.)