Johanna Larsson 发表评论文章指出，当前大语言模型的设计目标与人类学习的需求相违背 1。她认为，LLM 被训练为自动执行任务而非辅助学习，这种设计导向使用户容易过度依赖这些工具，进而丧失对基础概念的理解和实际操作能力 1。
为了保持有效的学习体验，Larsson 提出了若干限制 LLM 自主性的指导原则，包括限制代理权限、要求提供人类资源和明确指令等 1。她强调，"你越是抗拒让代理自动完成任务的诱惑，学到的东西就越多，即使这个过程令人烦恼或沮丧。" 1 这一观点指向了一个根本的设计困境：当前 LLM 的盈利模式与最优学习体验存在本质不兼容 1。
Johanna Larsson has published a commentary arguing that the design of current large language models fundamentally conflicts with human learning objectives 1. According to Larsson, LLMs are engineered to autonomously execute tasks rather than to support the learning process, which encourages users to become over-reliant on the technology and lose their grasp of underlying concepts and practical capabilities 1.
To preserve learning effectiveness and engagement, Larsson proposes several guiding principles for constraining LLM autonomy, including limiting agent permissions, requiring human-provided resources, and issuing explicit instructions 1. She emphasizes that maintaining this restraint—even when inconvenient or frustrating—yields better learning outcomes, noting that "the less you give into the temptation of the agent doing things, the more you learn, even when it's annoying or frustrating" 1. Larsson contends that the current profit-driven business models surrounding LLMs are fundamentally incompatible with optimal learning experiences 1.
评论
还没有评论，欢迎留下第一条。