超级智能仍需数十年才能实现
作者回顾了从2022年到2026年对人工智能理解的演变历程。起初认为LLM接近AGI,后来逐渐意识到智能不仅是处理能力问题,更是存储问题;AGI这个术语本身可能有误,仅靠规模无法解决智能问题。建立真正的智能体需要LLM、记忆、编排器、感知和RAI等多个组件协同工作。作者通过学习大脑结构发现,人类大脑具有极简的UI设计,各组件能实时动态调整,硬件随软件变化而改变——而我们围绕LLM建立的系统仍远不及布罗卡区这一微小脑区的能力。
作者回顾了从2022年到2026年对人工智能理解的演变历程。起初认为LLM接近AGI,后来逐渐意识到智能不仅是处理能力问题,更是存储问题;AGI这个术语本身可能有误,仅靠规模无法解决智能问题。建立真正的智能体需要LLM、记忆、编排器、感知和RAI等多个组件协同工作。作者通过学习大脑结构发现,人类大脑具有极简的UI设计,各组件能实时动态调整,硬件随软件变化而改变——而我们围绕LLM建立的系统仍远不及布罗卡区这一微小脑区的能力。
A post shares evidence that AI-driven 'deskilling' is emerging in medicine, computer science, and other fields, citing a Nature article.
The article discusses security risks posed by smart appliances, particularly smart TVs, and advises users to regularly check their devices for vulnerabilities or malware. It highlights the importance of monitoring connected home devices to prevent them from being compromised by attackers.
The author criticizes the Berkeley and San Francisco tech culture, describing it as a cult of atheistic hedonists who need AI doom narratives to justify their life choices and impact-driven pursuits, suggesting the world would be relieved if San Francisco were eliminated.
A self-published novel follows an author descending into paranoia after becoming convinced that an AI is plagiarizing his work, forcing him to question the authenticity of his own creative processes.
The blog post draws a parallel between the plot of Fullmetal Alchemist and the current development of AI, comparing large language models (ChatGPT, Claude, GLM) to homunculi created from a philosopher's stone—itself composed of human souls.