I wrote in @technologyreview.com recently that if AI summaries (e.g. Perplexity) impact click-thru rates for content—draining eyeballs and ad revenue—that could revive a messy debate over content bargaining. YouTube may be the canary in the coalmine.

www.technologyreview.com/2024/10/31/1...

Folks have different views, but mine is – governments should regulate narrowly, scrutinize for unintended effects, and promote grassroots innovation around open models in every future reform.

...and these interventions can disproportionately affect grassroots innovation in open models. They can (i) restrict the public release of capable models, (ii) require the impossible of open developers, or (iii) hold everyday developers to the standard of OpenAI and Meta.

But one implication is that governments are increasingly focused on model-layer regulatory intervention to solve all these problems...

And sometimes they represent genuine disagreements about what "really matters" in AI policy (a dispute that will intensify post-election)...

There are some 1,000 federal and state bills; the Executive Order was the longest in US history; and the AI Act ran to nearly 700 pages at one point. Yet many of these instruments are pulling in different directions. Sometimes these variations are natural...

There is essentially no consensus about why we're regulating AI, how to regulate AI, or whom to regulate in the tech stack...

Against the backdrop of a noisy AI culture war...