Ryan Greenblatt – Human level AIs might build runaway superintelligences by 2032

Ryan Greenblatt – Human level AIs might build runaway superintelligences by 2032 – Dwarkesh Podcast

Market views

Bullish

AILABS— AI labs are positioned to capture massive value through recursive self-improvement in AI R&D, potentially achieving 4-5 years of progress in a single year once AIs match top human experts. This could lead to an industrial explosion with AIs building out way more compute and developing technologies humans struggle to understand.
DATACENTER— Data center and compute infrastructure will see massive expansion as AI labs scale up training runs and AIs themselves begin building out compute capacity during the industrial explosion phase.
ROBOTICS— Robotics development is highly commingled with AI research progress, and human-level AI operating robots could enable rapid transformation of physical industries once AIs become capable enough at verifiable tasks like chip design and fab construction.

Bearish

ALTDATA— The importance of scaling human expert data labeling for AI progress is overstated. Algorithmic improvements and better data curation methods are more important drivers than spending on human expert labor, with compute-to-data spending ratios around 10-20 to 1.

Not investment advice.