Does intelligence need a hard cap?
Calls for a new kind of slowdown were the talk of The Curve. PLUS: Slinking back to X
During The Curve conference, leaders from top AI labs, nonprofit organizations, government officials, and journalists debated questions of economics, politics, and safety. A new push emerged to consider limits on the intelligence of large language models. This comes after the OpenAI-Hugging Face incident and OpenAI and Anthropic's progress toward recursive self-improvement, which could lead to fast releases and possible control loss.
Some speakers suggested limiting the intelligence of future systems to prevent superhuman intelligence. This idea, made in the Chatham House Rule, struck me as noteworthy. Limiting intelligence could involve restrictions on using frontier models for AI research, limiting available compute, or preventing deployment past a certain capability.
However, enforcing these restrictions is a challenge. Individual labs cannot impose such limits, nor can individual countries, and the US government actively opposes them. It remains unclear how close we are to actual danger, with lab leaders and the US government having differing views. The president's insistence on calling it "super intelligence" adds to the confusion.
Written by urgent.news from Platformer's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
