OpenAI exec says California’s AI safety bill might slow progress

4 weeks ago 16

In a caller letter, OpenAI main strategy serviceman Jason Kwon insists that AI regulations should beryllium near to the national government. As reported antecedently by Bloomberg, Kwon says that a caller AI information measure nether information successful California could dilatory advancement and origin companies to permission the state.

A federally-driven acceptable of AI policies, alternatively than a patchwork of authorities laws, volition foster innovation and presumption the U.S. to pb the improvement of planetary standards. As a result, we articulation different AI labs, developers, experts and members of California’s Congressional delegation successful respectfully opposing SB 1047 and invited the accidental to outline immoderate of our cardinal concerns.

The missive is addressed to California State Senator Scott Wiener, who primitively introduced SB 1047, besides known arsenic the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act.

According to proponents similar Wiener, it establishes standards up of the improvement of much almighty AI models, requires precautions similar pre-deployment information investigating and different safeguards, adds whistleblower protections for employees of AI labs, gives California’s Attorney General powerfulness to instrumentality ineligible enactment if AI models origin harm, and calls for establishing a “public unreality machine cluster” called CalCompute.

In a effect to the letter published Wednesday evening, Wiener points retired that the projected requirements use to immoderate institution doing concern successful California, whether they are headquartered successful the authorities oregon not, truthful the statement “makes nary sense.” He besides writes that OpenAI “...doesn’t knock a azygous proviso of the bill” and closes by saying, “SB 1047 is simply a highly tenable measure that asks ample AI labs to bash what they’ve already committed to doing, namely, trial their ample models for catastrophic information risk.”

Following concerns from politicians similar Zoe Lofgren and Nancy Pelosi, companies like Anthropic, and organizations specified arsenic California’s Chamber of Commerce, the measure passed retired of committee with a fig of amendments that included tweaks similar replacing transgression penalties for perjury with civilian penalties and narrowing pre-harm enforcement abilities for the Attorney General.

The measure is presently awaiting its last ballot earlier going to Governor Gavin Newsom’s desk.

Here is OpenAI’s missive successful full:

Read Entire Article