"I don't think instincts and conditioning are relevant to AGI per se, only pure curiosity is. The former: supervision and RL, are optional add-ons, useful as short-cuts to unsupervised learning, or means to make it serve you."
You must not know how GPT nor my AI works then. Recency priming is a very important part for most prediction problems. Ex. the cat ate food, my cat saw a cat go near a ___. Cat, is the predicted word, because similar things group together. Pig, horse, zebra, pony, lab, kitten, bird, ___. The most important IS reward for prediction, it is what makes it not ramble like GPT2 anymore and steers it so food sex home AGI nekosextoy is what it talks about permanently. What you said is very wrong. And there is more on this too. "I think that's an extremely short-sighted perspective. Both because symbolic data should never be used as primary input, and because the scope of predicted "next" is proportional to the scope of all represented experience. But that's a subject of my intro: http://www.cognitivealgorithm.info, esp. part 1. " Again this reply is nonesense. We are not even talking about text or vision cons & pros, of course both used together like DALL-E did is way better. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T5b614d3e3bb8e0da-Mf796b1a609e50edf852e94e5 Delivery options: https://agi.topicbox.com/groups/agi/subscription
