Paris Call for AI Safety and Security sees six governments and ten companies, including OpenAI, Anthropic and Microsoft, pledge fail-safe mechanisms to shut down models. The post calls for states to set thresholds and a January deadline and to build capacity for an internationally coordinated slowdown.
Key points
Six governments and ten companies commit to fail-safe mechanisms to shut down models, including during training and internal use.
States should set thresholds by January and build capacity for an internationally coordinated slowdown in case of emergency.
Sources
Hard Cases for AI Behavioural Design paper released
Hard Cases for AI Behavioural Design paper released, outlining hard cases for AI behaviour in consequential institutions and a systematic walkthrough of key tradeoffs. The paper is co-written by the poster and @willmacaskill.
Key points
Title: Hard cases for AI behavioural design
Authors: the poster and @willmacaskill
Sources
- Original post
- Hard cases for AI behavioural design - Linked page