《麻省理工科技评论》指出,当前的大语言模型(LLM)并不进行真正的推理 1。研究员Nick Cammarata撰写了一篇26,000字的病毒式博客文章,其观点颠覆了AI世界,并受到AI安全领域人士的庆祝 1。在AlphaGo对战李世石十周年之际,文章提到当今人工智能仍未触及当年的获胜机制 1。
随着个人AI助手的兴起,“代理型网络”已经到来,互联网不再仅服务于人类 1。此外,今年夏天的AI炒作(如AGI和新能力声明)在审视下迅速崩塌 1。
MIT Technology Review’s subscription content centers on critical assessments of current artificial intelligence capabilities, particularly challenging the notion that large language models possess true reasoning skills 1. The publication notes that despite the tenth anniversary of AlphaGo’s victory over Lee Sedol, contemporary AI systems have yet to approach or replicate the specific mechanisms that secured that historic win 1. This skepticism extends to recent industry hype, with claims regarding Artificial General Intelligence and novel capabilities reportedly collapsing under scrutiny this summer 1.
The reports also spotlight the emergence of the "agentic web," a paradigm shift where the internet increasingly serves personal AI assistants rather than human users alone 1. A key point of discussion involves researcher Nick Cammarata, whose 26,000-word viral blog post has significantly disrupted prevailing views in the AI community 1. Cammarata’s arguments have been celebrated by figures in the AI safety sector for overturning established assumptions about model capabilities 1.
评论
还没有评论,欢迎留下第一条。