Dario Amodei, CEO của Anthropic, phản hồi về những lời đồn rằng công ty ông ủng hộ lệnh cấm của chính phủ Mỹ đối với các mô hình AI mở trọng lượng của Trung Quốc, khẳng định Anthropic không bao giờ ủng hộ lệnh cấm này.
Anthropic’s Stance on Open-Weight Models and AI Bans
Clarifying Anthropic’s Position
Dario Amodei, founder and CEO of Anthropic, addressed the rumors and concerns within the industry regarding Anthropic’s alleged support for banning open-weight Chinese models or open-weight models in general. He emphatically stated that Anthropic has never advocated for such a ban, dispelling any doubts about their position.
Nvidia’s Open Letter and Industry Debate
The discussion gained momentum after Nvidia’s CEO, Jensen Huang, posted an open letter on X, signed by numerous AI companies, urging policymakers to refrain from imposing broad restrictions on open-weight AI models. While China was not explicitly mentioned, the industry debate revolves around accusations that Chinese AI labs are advancing by stealing intellectual property from American companies. The method of distillation, where one AI model learns from another through prompts, is a point of contention.
Amodei’s Perspective on Open-Weight Models
Amodei differentiated between open-weight models and the perceived threat from China. He believes that open-weight models without dangerous capabilities are beneficial to the public, offering value at a minimal cost. Amodei’s fears about AI are centered around authoritarian governments building superior models for military dominance or internal repression, with the Chinese Communist Party being the most capable among them.
Concerns about AI-Enabled Threats
The Anthropic CEO expressed concerns about AI enabling biological attacks, in addition to cybersecurity threats. He considers open-weight models more dangerous in these scenarios due to the challenges of implementing guardrails and monitoring their usage. Amodei referenced a UK AI Security Institute report, highlighting the irreversible nature of releasing open-weights models, contrary to the open-source community’s belief in the benefits of decentralized control of powerful AI models.
The Debate Continues
This response from Amodei adds a nuanced perspective to the ongoing debate within the AI industry. While some advocate for open access and collaboration, others, like Amodei, emphasize the potential risks associated with unrestricted AI development, especially in the context of geopolitical tensions and the rise of authoritarian regimes.
