Article title: On Anthropic's IPO Day, Former Researcher Coxon Calls Outside Its Headquarters for a Ban on AI Recursive Self-Improvement
Body:
- Day/Night Follow system Light Dark
On Anthropic's IPO Day, Former Researcher Coxon Calls Outside Its Headquarters for a Ban on AI Recursive Self-Improvement
IT之家 November 20 news, according to Business Insider, in September this year, former Anthropic researcher Jacob Coxon left his job due to concerns about AI development and sparked widespread panic about AI technology in the United States. On the 19th local time, as Anthropic's shares began trading publicly for the first time, Coxon once again called on the public to pay attention to AI safety issues.
This time, in an interview, he stressed: "Whether they buy the stock or not doesn't matter at all. We must stop recursive self-improvement."
IT之家 note: "Recursive self-improvement" refers to having advanced AI systems participate in training the next generation of AI, thereby driving continuous improvements in AI capabilities.
That day, Coxon and dozens of protesters gathered outside Anthropic's San Francisco headquarters, demanding a ban on the technology in question.
Protesters held signs and chanted slogans, and the scene also featured speeches by Coxon, former OpenAI researcher Daniel Kokotajlo, and YouTube creator Rob Miles, while members of the Stop AI organization who arrived later blocked the entrance to the office building.
An independent report recently released by AI safety research organization METR found that AI is helping Anthropic accelerate its R&D progress. Anthropic also acknowledged in its IPO prospectus that letting AI fully autonomously design the next generation of models without oversight is not safe, and it is currently considering policy changes to ensure that the development of frontier AI models remains under human control.
Coxon believes these measures alone are not enough. In his speech, he called on Anthropic to lobby the government and cooperate with other companies to push for a comprehensive ban on recursive self-improvement. "At this point, companies themselves are the problem. In 2027, only a few places in the world will be able to achieve recursive self-improvement, and this is one of them."
Both Coxon and Kokotajlo mentioned that they often hear AI researchers privately express concerns about the risks of the technology. Kokotajlo described it: "This kind of thing is exactly what those mad scientist-style supervillains in comics would do."
Related Articles
- "The largest IPO in history": Anthropic closes up 17% on its first trading day, with a market value of 1.27 trillion USD
- Anthropic completes the largest IPO in history, with an IPO valuation of 1.5 trillion USD
- The largest IPO in history brings a new round of wealth-creation mythology: Anthropic's listing may create thousands of millionaires
- Anthropic's IPO reaches a record 107.3 billion USD, with gray market trading indicating an opening gain of more than 20%
- The largest IPO in history: Anthropic priced at 83 USD per share, expected to raise 94.4 billion USD, surpassing SpaceX
- Sources say Anthropic is considering expanding its IPO, with retail subscription interest approaching 100 billion USD
Websites under Soft Media: IT之家 最会买 - 返利返现优惠券 iPhone之家 Win7之家 Win10之家 Win11之家
Software under Soft Media: 软媒手机APP应用 魔方 最会买 要知