The case for regulating harmful AI uses rather than the technology itself

The discussion distinguishes restrictions on dangerous AI uses from restrictions on a general-purpose technology. One commenter worries about lasting control by a few big labs and wants more open-source or open-weight models, eventually allowing people to run AI on their own devices.

A reply argues that specific uses and threats are easier to regulate, citing proposed know-your-customer rules for cloud providers. It also doubts that FLOP counts—counts of computing operations—could underpin workable regulation. The original commenter agrees that people should be accountable for harm they cause with AI and supports reasonable measures to identify those responsible.