I'm the same age as Larry and started working in machine learning in the early 90s, and to me, it seemed pretty obvious at the time that AGI would ultimately ne

I'm the same age as Larry and started working in machine learning in the early 90s, and to me, it seemed pretty obvious at the time that AGI would ultimately need a large corpus of text (and video) to be useful. What's kind of funny is the models I worked with at the time- hidden markov models of (biological) text anticipated the validity of this approach, although by design, they don't work with long contexts.