Risks and Windfall Clauses in AI

22min Snip

00:00

Play full episode

Summary

Transcript

Episode notes

The chapter discusses the risks associated with advancing AI, including super-intelligence, control of power, and human disempowerment. It explores the effectiveness of windfall clauses in dangerous AI scenarios. The conversation also touches on AI agents' motivations and parallels to genocide.

Read the full transcript here.

Will large language models (LLMs) replace journalists any time soon? On what types of writing tasks do LLMs outperform humans? Have the US news media become less truth-seeking in recent decades? Or is truth-seeking behavior merely an aberration from a norm of propagandizing? How should we redistribute economic surplus from AI? Have any AI companies committed to a Windfall Clause? Instead of bothering to negotiate with us, wouldn't a superintelligent AI be able to get much more done by first wiping us all out? What are some subtler or less-well-known ways subscription models reshape incentive structures for journalists? Why is collective action so hard?

Dylan Matthews is a senior correspondent at Vox, where he cofounded Future Perfect, a section devoted to exploring ways to do good. He writes frequently about economics, philanthropy, global health, and more. You can email him at dylan@vox.com.