4 ms·
> trained on 12 million hours of speech Humans train to recognize speech on much smaller datasets. If a small human is awake 16 hrs/day, that amounts to maxim
by codedokode 4y ago
> trained on 12 million hours of speech
Humans train to recognize speech on much smaller datasets. If a small human is awake 16 hrs/day, that amounts to maximum 5840 hrs/year or 58400 hrs per 10 years. Why do mathematical models use more data and produce lower quality results? Is it because they don't understand the meaning of words?
- verbify 4y agoThe dataset humans are using does not just include the audio stream but also visual context. E.g. people pointing at things while talking about them.
- tgv 4y agoBlind people...
- nr2x 4y ago…or deaf.
- deskamess 4y agoFun/interesting question... do you think blind people gesture when speaking?
- tgv 4y agohttps://news.uchicago.edu/story/blind-adults-gestures-resemble-those-other-native-speakers https://news.uchicago.edu/story/blind-adults-gestures-resemb... It's not from a department I'm a fan of, but they do know how to catalog gestures.
- b0afc375b5 4y agoTo be fair, humans have the benefit of millions (billions?) of years of evolution.
- vkazanov 4y agoMy personal explanation is how large ML models are stateless while the growing brain is a (very) stateful thing. Stateless imitation will never be able to fully replicate a stateful system. We evolved into having these brains that learn in stages, when the early years are responsible for core functions (walking, running, talking, etc). There's a lot of input (all the senses) that feeds learning, while in later years we mostly just take for granted whatever we learnt as children.
- wasabi991011 4y ago>that amounts to maximum 5840 hrs/year or 58400 hrs per 10 years But no human can understand 300+ languages, especially not at 10 years old. Still a very approximate comparison, but if you multiply your upper bound by 300, that gives 17.5 million hours, so more than what was used to train.