I like this article. It's a fairly accessible and intuitive description of how an AI learns through pattern recognition, and it offers lessons, I think, in how humans learn. "Every time it makes a guess, we can measure exactly how wrong it was. But there's more to it than simple word prediction. The model also learns to understand context in increasingly sophisticated ways. For instance, when it sees 'The cat sat on the mat because...' it's not just learning about cats and mats - it's learning about causality, the relationship between tiredness and sitting, and the typical behaviours of animals." This (to my mind) is why direct instruction focusing on specific content is not the best approach - by eliminating all extraneous context in the name of cognitive load it eliminates the possibility of extraneous learning (like causality and how animals behave), and this really is the most important part of the learning.
Today: 0 Total: 279 [Direct link] [Share]
