Stefan has 2 big problems: #1 Matt is very educated, #2 I also agree hand crafted rules is the wrong path. The experts are against you, Stefan! :-). Give it up. Unsupervised Learning is the ultimate way to soak up massive data, to power the AI short cut finding ability I spoke about by using analogies from all domains. Yes we must hardwire the AI a bit, but we just talk to it afterwards, no hardwiring each memory syntax or semantics, the world does that on its own! Oof!
I already explained to you above Stefan how it works is this: "The cat ate food on the lawn outside" Cat ate what? Answers: Food/lawn/outside "The cat ate food on the lawn outside" You're just looking for the best matches in the story heard so far using Probability. GPT-2 is all about this, so is seq2seq and word2vec and BERT and Data Compressors. They GRAB the next word to predict - by finding in memory a MATCH/ES from recently seen text/data and the end of it has the word/letter that entails that piece; the answer. Frequency is used to choose which prediction that entails. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T21bdc2c440c86db7-Me0770d88568bd05908324f0a Delivery options: https://agi.topicbox.com/groups/agi/subscription
