Top AI executives call for a slower pace for the most powerful AI models

In short: Some of the biggest names in AI are publicly urging the industry to slow down the development and release of the most powerful AI systems because safety work may be falling behind.

What's going on

Several top AI leaders, including Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and xAI owner Elon Musk, have supported calls to slow the pace of work on “frontier” AI models. Frontier models are the most capable systems, like the ones at the front of the pack.

The point is not to stop AI or ban everyday products. The proposals focus on pacing the training and release of the most powerful models, similar to how a car company might delay shipping a faster car until better brakes and seatbelts are tested.

Supporters of a slowdown say they worry about “loss of control.” They also worry about AI systems that can improve themselves over time (like a machine that keeps rewriting its own instruction manual), and about AI agents that can take actions on their own. An AI agent is software that can plan and carry out tasks, like booking tickets or writing and sending messages, without a person approving every step.

What to watch

Ideas being discussed include releasing frontier models less often, requiring extra approval for systems that could help with cyberattacks or biological harm, letting independent monitors review safety practices at AI labs, and negotiating international limits. The debate has also triggered pushback in the US and China, where officials worry that slowing down could change competition between companies and countries. Separately, some researchers argue that older methods for making AI better may be hitting limits, but that is a different discussion from the safety-driven calls by executives.

Source: NYTimes