推文

@Michaelzsguo · 2026-10-17 02:00

Do these two images look like the same thing to you? Of course not. But to a neural network, they are effectively identical. This is from a 2019 MIT study showing how radically machine perception can differ from ours. Nick Cammarata and Dean Ball are onto something profound: Machine representations aren't necessarily meaningless just because humans can't interpret them. Perhaps the key to understanding AI isn't forcing its internal representations into human concepts, but discovering the structure and meaning they have on their own terms. What looks like noise to us may be perfectly meaningful to a machine. (MIT research: https://news.mit.edu/2019/differences-between-deep-neural-networks-and-human-perception-1212?utm_source=chatgpt.com)

曝光 37 · 评论 1 · 点赞 0 · 书签 0