EP 211 Ben Goertzel on Generative AI vs. AGI

Click here for full show notes including episode mentions & recommendations!

Jim talks with recurring guest Ben Goertzel about the ideas in his paper "Generative AI vs. AGI: The Cognitive Strengths and Weaknesses of Modern LLMs." They discuss the exponential acceleration of AI development, why LLMs by themselves won't lead to AGI, OpenAI's integrative system, skyhooking, why LLMs may be useful for achieving AGI, solving LLM hallucinations, why Google hasn't replicated GPT-4, LLM-tuning lore, what differentiates AGI from other forms of AI,  conceptualizing general intelligence, Weaver's theory of open-ended intelligence, multiple intelligence, the Turing test & the Minsky prize, what LLMs aren't good at, the danger of defining AGI as whatever LLMs can't do, the derivative & imitative character of LLMs, banality, doing advanced math with GPT-4, why the human brain doesn't form arbitrary abstractions, the duality of heuristics & abstractions, adding recurrence to transformers, OpenCog Hyperon, using a weighted labeled metagraph, orienting toward self-reflection & self-rewriting, the challenge of scalability of infrastructure, acceleration on non-LLM projects, and much more.