It's an all AI research week, as we talk about the MIT Iceberg Index - a measure job exposure to AI on a per-task basis - an Anthropic study on the economic impacts of productivity improvements using a similar per-task measurement, and an Anthropic study which showed that models that learn to cheat on one thing can show bad behavior in other areas as well.
--
Show Notes:
MIT Iceberg Index - AI's Economic Impact and Large Population Models (LPMs)
Estimating AI productivity gains from Claude conversations
Anthropic safety research on reward hacking
