>>todsac+(OP)
I would argue that Lenat was at least directionally correct in understanding that sheer volume of data (in Cyc's case, rules and facts) was the key in eventually achieving useful intelligence. I have to confess that I once criticized the Cyc project for creating an ever-larger pile of sh*t and expecting a pony to emerge, but that's sort of what has happened with LLMs.
>>vannev+14 https://ai-2027.com/ postulates that a good enough LLM will rewrite itself using rules and facts... sci-fi, but so is chatting with a matrix multiplication.
>>baq+3j
How will the rules and facts be connected? By some discrete relationship?
This stuff only works for math, and is the basis for the bitter lesson.
Intelligence is compression, and this is the opposite of that