Sometimes I think of the LLM and RL schools of AGI as two theories of intelligence. Is intelligence about being able to answer any question, or being able to achieve any goal? Now that we’ve conquered the former, attention has turned to the latter mission.
Exploring the internet together should be like exploring a vast old library with your friends. Wandering down different shelves, skimming the pages that catch your eye, and occasionally one of you hollers in a whispering voice, "come check this out!"
Scraping the Social: “We are unknown to ourselves—and with good reason.” Friedrich Nietzsche—“Even the retards are starting to figure it out.” (comment)—“In data we trust.” Priceonomics—“The Internet fails to scale gracefully.” Chris Ellis—“I want to be surprised by my own bot”—“There is a crack in everything. That’s how the light gets in.”... See more
the worst of all possible worlds: generative AI manages to pollute the internet with cheap synthetic data, manages to make being a human artist / creator harder