知名开发者Rich Hickey在2026年10月12日发表文章，提出大型语言模型（LLM）不应被称为工具，而应理解为自动化系统1。Hickey认为，真正的工具需要使用者能够施加持续控制，但LLM的运作方式是用户提出请求后系统自主行动1。他指出，使用LLM进行写作相当于放弃对代码的直接编写权1，使用者仅在适应算法而非掌握真实世界知识1。
这一观点随即引发回应。持不同立场的作者认为，尽管LLM具有不可预测性，但仍可被视为工具，因为工具的定义并不要求完全可控2。该作者指出，编译器和电子表格等自动化工具依然被广泛认可为工具，文本生成工具也应如此归类2。在该作者看来，关键区别在于操作者是否有效运用工具并充分理解其原理，而非LLM整体是否属于工具范畴2。使用LLM同样需要操作者具备明确意图、建立反馈循环、进行适当指导和拥有专业知识2。
Renowned developer Rich Hickey published an essay on October 12, 2026, arguing that large language models should not be classified as tools but rather understood as autonomous systems 1. Hickey contends that true tools require fundamental characteristics including controllability, feedback loops, and intentional direction by the user 1. He asserts that LLMs operate fundamentally differently: "You don't direct the action of an LLM—you make requests and it acts on its own" 1. According to Hickey, using an LLM for writing amounts to abdicating the task rather than wielding a tool, since users adapt to the algorithm rather than mastering their craft 1. He further warns that knowledge acquired through LLM use remains precarious, noting "You are only a software update from everything you've 'learned' being rendered obsolete by the latest algorithm" 1.
The thesis sparked immediate rebuttal from another commentator, who maintains that LLMs merit the designation of tools despite their unpredictability 2. This respondent argues that the definition of a tool does not require complete control, and points to existing automated systems—compilers and spreadsheets—that continue to be recognized as tools despite their inherent complexity 2. The counterargument emphasizes that proficient LLM use demands the operator's intent, feedback interpretation, guidance, and domain expertise, much like any skilled tool operation 2. Additionally, the response critiques the artificial intelligence industry's anthropomorphic language, noting the problematic deployment of phrases such as "go rogue," "escape containment," and "conspire" 2.
评论
还没有评论，欢迎留下第一条。