Ben Brooks

Nearly every AI firm is invoking frontier risk to justify a persistent gap between open and closed models. But this application of the precautionary principle deserves more scrutiny than it gets.

Restricting access to useful technology—models that will, in their developers' own words, transform the economy—shouldn't be our primary response to uncertain risks.

Check it out at @aif_media! There are already a bunch of reasons firms might not release their best models openly: cost recovery, competitive pressure, anxious investors. But we should be skeptical of efforts to freeze open-source behind the frontier under the guise of risk.

The spluttering indignation over the data center backlash is bewildering to me. Frontier labs and their thinkfluencers have spent ~2 electoral cycles telling anyone who cares to listen that frontier capability will at best upend the economy and at worst kill everyone.

Open-weight deployers value cost efficiency *and* intelligence. They also value transparency, composability, modifiability, security, privacy, and operational independence.

A fun night at @github HQ fighting for the proposition that democratizing AI means owning, not just renting, intelligence.

In a ¡shocking! turn of events, open source ultimately prevailed in the Open Source AI Week Debate 😂

Thanks for bringing us together @matthew_d_white, @ekang426, Margaret Tucker, and crew. Glad to link arms with @istoica05 and @percyliang to defend openness—and big kudos to Cheryl Cheng @M12vc, @m_bourgon, and @DavidRBellamy for taking on the GH audience with aplomb!

Two features stand out to me in China's proposed export controls for model weights:

1. They define granular thresholds for covered systems that focus on verifiable or semi-verifiable precursor indicators (e.g. attack success rates).

For exotic (i.e. unrealized) AI risks, I am generally more optimistic about narrowly defined triggers for regulatory intervention than broad duties of care.

2. Existing open-source technology is exempt, and future open-source releases are subject to automatic licensing.

Loosely speaking, this seems to resemble the "notification-but-not-authorization" requirements that already apply to open releases of non-standard cryptography in the US.

To be clear, export controls for general-purpose weights are directionally wrong. But there is a refreshing level of concreteness and seriousness in these proposals that is totally absent in US discourse over export controls.