Dwarkesh Patel Video posted an update
I thought superintelligence was a decade away. Not anymore
Why it mattersThe creator explains why he has changed his estimate for when superintelligence could arrive, discussing AI accelerating AI progress, automating researchers, and the prospect of AI as a meta-learner. He also lays out objections to recursive self-improvement, including continual learning, verifiable proxies, data, and experiment compute.
Discuss: Which challenge to recursive self-improvement-continual learning, verifiable proxies, data, or experiment compute-seems most important to resolving the creator's argument?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.