权威AI批评家Gary Marcus再次发声，直指LLM根本不可靠，加上特朗普政府与Anthropic的对峙，这场AI安全争议你必须了解。
Gary Marcus重申LLM不可靠的观点，称无法被驳斥。特朗普政府要求Anthropic确保Fable 5的护栏不能被绕过，但安全专家表示不可能完全阻止规避。Marcus认为这是生成式AI的普遍问题，而非Anthropic一家的问题。WIRED报道引述了相关官员和专家的意见。
For years I have been saying that LLMs are not reliable enough to be sound. That simple claim — wh...
For years I have been saying that LLMs are not reliable enough to be sound. That simple claim — which has never been refuted — is suddenly at the center of the universe. What Trump asks cannot be done. NO current system can completely resist circumvention. Either we curtail LLMs until we find a better technology, or we live with consequences. This is not an Anthropic problem, it’s a Generative AI problem. WIRED @WIRED Trump administration officials tell WIRED that if Anthropic wants to rerelease Fable 5, it will need to ensure the model's guardrails can't be circumvented. Security experts say that can't be done. wired.com/story/the-whit… 🔗 View Quoted Tweet 💬 8 🔄 24 ❤️ 79 👀 7839 📊 17 ⚡