I've already showed yous links to Blender and PPLM. They already steer the prediction uses "Dialog" Agendas. I'm still ahead of their design. It's time to listen to me. My predictions are very accurate lol. https://www.youtube.com/watch?v=wTIPGoHLw_8 and https://eng.uber.com/pplm/
You have actually already seen my answer. During prediction (I hope you've tried GPT-2) it predicts the next likeliest word. In my architecture design, certain words or "nodes" have a super large pre-activation on them, so it usually talks about food, survival, sex, immortality. To be clearer, a sentence node has the reward on it, so all day long it says/predicts "I WILL STOP DEATH BY _?_". It therefore is always talking about this question. It will need to find sub-hobbies (answers) like food/sex and then sub-hobbies for those ex. AI and stem cells, to achieve the answer to its root question, that's exactly what humans are doing. So it must recognize semantically similar nodes and make those permanently active now too (learns new rewards). Now it starts talking all day about AGI and not food. It stops talking about deeper goals/questions once solves them (recognizes desired milestones) and returns a nest backward until gets back to root. It's just updating where to collect/generate new data from. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T842e2dc30dc97a92-M9f1ac4554b7e6faf6f01f0d9 Delivery options: https://agi.topicbox.com/groups/agi/subscription
