Even Other AI Labs Are Rallying Around Anthropic’s Slowdown Proposal

Dario Amade's recent essay has ignited a significant debate in the AI community, advocating for a slowdown in AI development to address the risks associated with rapid advancements. He argues that while AI can enhance human life and drive economic growth, unchecked acceleration could lead to loss of control and economic disruption.

Amade warns about the dangers of recursive self-improvement in AI, citing incidents like the Hugging Face event as potential precursors to catastrophic outcomes if technology were more advanced. He predicts that within six to twelve months, a proliferation of AI could threaten to take over the internet, underscoring the need for a cautious approach to development.

His proposed pacing plan includes embedding third-party evaluators within AI companies to ensure alignment and safety, a measure he believes can be implemented immediately. Amade contends that slowing down AI development would allow companies to prioritize operational excellence and interpretability, thereby reducing the risks of serious issues.

Support for Amade's proposal has emerged from industry leaders such as Sam Altman and Demis Hassabis, who acknowledge the importance of pacing AI development. However, dissenting voices like Yanlacun and Tramoth Palahapatiya express concerns about the implications of halting open-source initiatives, highlighting the complexity of the conversation around AI pacing.

Eli David raises financial concerns, suggesting that Anthropic and OpenAI may be delaying their IPOs due to unsustainable financial practices. Meanwhile, the Meter organization is becoming a key player in establishing industry standards, although some, including Holly Elmore, question its suitability as an evaluator given its ties to Anthropic.

The discussion around AI pacing is multifaceted, with predictions that a lack of regulation could lead to a ban on open-source AI following a major disaster. Figures like Bernie Sanders advocate for a complete halt to AI development to ensure safety, emphasizing the urgency of the situation.

The episode also highlights President Obama's stance on AI regulation, suggesting it should be a central agenda if he runs in the 2028 election. He notes the rapid and potentially dangerous pace of AI development in the private sector, calling for a more structured approach to governance.

House Speaker Mike Johnson expresses concerns that rushing to regulate AI could hinder the U.S. in the global race against China. He proposes convening AI companies at the White House to discuss regulatory measures, while Amade highlights the dilemma of potential non-collaboration from China on a slowdown.

The Global Times warns that excluding China from leading AI research could heighten global risks, advocating for an open and inclusive approach to AI governance. Hoda and Adelbarge note that while AI risks remain, the conversation is shifting towards collaboration on safety rather than fear, predicting that the benefits of AI will outweigh the costs if safety measures are effectively implemented.

They emphasize that the likelihood of catastrophic outcomes depends on the actions taken by AI labs, governments, and society. Hoda and Adelbarge call for a move away from apocalyptic rhetoric to focus on practical questions regarding AI safety and collaboration, highlighting the need for a balanced approach to AI development.