David Sacks on X: Some thoughts on Dario's post

Some thoughts connected Dario’s post: 1. Dario does not really reside Gavin Baker’s relationship of what he said – thing he could easy contradict if it were inaccurate. 2. Dario claims his critics unrecorded successful a “bubble” wherever each regularisation equals regulatory capture. He calls this an overly simplified position and notes that “Many group extracurricular this bubble deliberation of regularisation arsenic thing that constrains firm powerfulness and benefits mean people.” This statement is simply a straw man. Of people treating each regularisation arsenic seizure would beryllium overly simplified – but almost nary 1 holds that view. I person many times based on for beardown antitrust enforcement to support industries competitive, particularly Big Tech. If Anthropic continues toward monopoly aliases duopoly status, I would beryllium among the first to request those rules apply. 3. Regulatory seizure is not vague aliases successful the oculus of the beholder. Nobel laureate George Stigler defined it arsenic regularisation acquired by an manufacture and designed and operated chiefly for its benefit. Stigler challenged the accepted position that authorities regularisation arises from a benevolent authorities protecting the nationalist from marketplace failures. Rather, manufacture groups person concentrated stakes and move resources into influencing regulators, whereas the public’s liking is diffuse and unorganized. The revolving doorway betwixt companies and the agencies that modulate them compounds the problem. Anthropic understands these dynamics: it has hired aggregate elder Biden AI-policy officials and built a important government-affairs cognition positive a web of aligned organizations to push its preferred frameworks astatine authorities and national levels. 4. Dario has consistently pushed for a caller national agency to reappraisal and o.k. frontier models anterior to merchandise – a connection framed variously arsenic an “FDA for AI,” an “FAA for AI,” and astir precocious a “FINRA for AI.” I telephone it a “DMV for AI” because a reappraisal process modeled connected the FAA aliases FDA (which takes years) aliases FINRA (which issues rules for a staid manufacture wide seen arsenic protecting incumbents) will create agelong queues arsenic AI models hold for testing and approval. This process will only go much labyrinthine arsenic rules accumulate to forestall theoretical harms. This would handicap the U.S. comparative to China, which will not adopt the aforesaid constraints. It would besides undermine Anthropic’s ain business model, whose pricing powerfulness depends connected remaining up of unfastened models. Whatever Dario states today, it is difficult to judge the institution would simply judge outcomes that erase that advantage. 5. Anthropic is connected way to go 1 of the astir valuable companies successful history, pinch the resources to navigate immoderate support process and style the rules while competitors wait. Dario wants unfastened models nether heavier scrutiny – he has called them vulnerable successful Senate testimony, criticized them for not being centrally monitored aliases withdrawn, and linked them to IP theft. He says he has ne'er sought a ban, but he could execute a akin consequence by insisting that identical rules use to some unfastened and closed models. The U.S. risks becoming an land of costly closed models while the remainder of the world races up pinch broader choice. 6. Dario acknowledges that AI is structurally centralizing but attributes this chiefly to chips and scaling laws. Access to compute matters, but the deeper consequence is who decides which capabilities are disposable to whom. His preferred pre-deployment testing and FAA/FINRA-style oversight would spot that gatekeeping powerfulness successful a national bureaucracy moving hand-in-glove pinch a mini number of frontier labs – reinforcing centralization alternatively than countering it. 7. The 2nd portion of Dario’s station assumes we person amnesia astir Anthropic’s well-orchestrated campaigns hyping AI fears. His May 2025 declare that AI would swipe retired 50 percent of entry-level knowledge jobs wrong 5 years still lacks supporting grounds 15 months later. Similarly Anthropic breathlessly promoted its heavy contrived “blackmail” study connected 60 Minutes. Yet Dario blames nationalist negativity connected a long-standing nonaccomplishment of spot successful institutions alternatively than his ain messaging. 8. These narratives person done much than thing to style nationalist fear. People are near asking the aforesaid mobility Mark Zuckerberg posed: why title to build a early you picture successful specified antagonistic terms? Thomas Sowell’s "The Vision of the Anointed" captures the mindset – elite intellectuals convinced that only they are enlightened capable to power the outcome. As Zuckerberg notes, concentrating powerfulness successful the hands of an enlightened fewer has seldom produced the promised results; the practitioners move retired to beryllium little enlightened successful believe than successful self-conception. 9. Gavin Baker summarized the disagreement cleanly connected our pod: Dario believes frontier AI is excessively powerful to distribute; we judge it is excessively powerful to centralize. Dario appears to believe, sincerely, that information and advancement are champion served by centralizing authority successful a matrimony of firm and authorities power. The weight of quality history gives america logic to fearfulness that outcome.