So, the basic idea would be to have the two opposite parts: 1. the trained ANN (not necessarily LLM, but able to produce source code, probably improving over time) 2. the model of the world coded in some programming language (high level or low level, not sure about this)
ANN would program the source code by which the AI would communicate to the outer world. ANN would be merely a service which would constantly update the source code of its behavior. The source code would be the main point of AI behavior, with the model of the world programmed within. The starting point would subsume some grade of startup instinct being aware of "outer world -> AI input -> ANN -> source code -> AI output -> outer world" cycle. AI would constantly learn, improving its abilities over time. When the source code doesn't compile, ANN input is considered not compatible with the current code. There could be more than one version of world model, maybe intertwined one within another, just for testing which model fits the best. I guess there are still some unworked out details, but I think it should work. What do you think? ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/Tef2462d212b37e50-M5a87f7162f9244aea36b8ae7 Delivery options: https://agi.topicbox.com/groups/agi/subscription
