文章摘要
文章指出长时间使用大语言模型(如Claude、Codex)会让人精疲力竭,主要原因是用户疲劳会导致提示词质量下降,而休息后思路清晰时能更快找到解决方案。作者认为问题不在于模型本身,而在于使用者的精神状态。
文章总结
标题:大语言模型让人精疲力竭的真相
作者Tom Johnell在文章中分享了他与Claude、Codex等大语言模型长时间交互后的疲惫体验。通过自身经历,他总结出两个核心问题:
- 疲劳导致的恶性循环
- 随着精神疲劳加剧,提示词质量明显下降
- 常见错误包括:遗漏关键信息、频繁打断模型响应
- 这种状态下的交互往往事倍功半
- 缓慢的反馈循环
- 处理大型文件时,每次调整都需要耗时重新解析
- 上下文窗口很快被占满(接近98%容量)
- 导致模型要么变"笨",要么开始"假装理解"
应对策略: 1. 及时识别疲劳信号:当编写提示词失去乐趣时,就是该休息的信号 2. 警惕需求外包陷阱:不要指望AI填补未深思的需求空白 3. 优化反馈循环: - 将"缩短反馈时间"作为首要优化目标 - 借鉴TDD(测试驱动开发)理念 - 提供明确成功标准,让AI自动优化代码路径
核心洞见: 当使用LLM感到疲惫时,可能是"技能问题"。需要学会识别疲劳状态,避免需求外包的诱惑。如果编写完美提示词不再带来成就感,就该暂停或重新思考问题本质。对于缓慢的反馈循环,应该主动将其转化为待解决的优化问题。
(编辑说明:删除了原文中的网站导航栏、订阅信息等非核心内容,保留了作者的核心观点和具体案例,对技术术语如TDD作了必要保留并添加简要说明,确保专业读者能理解的同时也兼顾可读性。)
评论总结
评论内容总结
1. 调试与评估的挑战
- 观点:调试和评估AI代理的工作过程可能令人疲惫,需要不断检查其思考过程。
- 引用:
- "I think the exhausting part is more probably more tied to the evaluation of the work the agent is doing, understanding its thought process..." (chalupa-supreme)
- "Reviewing LLM output requires constant context-switching between 'what does this code do' and 'is this what I actually wanted.'" (stainlu)
2. 认知负担与任务切换
- 观点:同时运行多个代理或频繁切换任务会导致认知负担。
- 引用:
- "I’m much more likely to have 2-3 agent sessions running in parallel, resulting in constant switching which is very mentally taxing." (simonw)
- "You don’t want to sit waiting for it, but you don’t want to context-switch across more than a couple contexts yourself." (veryfancy)
3. LLM的不可靠性
- 观点:LLM输出不稳定,需要频繁纠正,类似指导一个不学习的新手。
- 引用:
- "LLMs do not actually make anything better for anyone. You have to constantly correct them." (razorbeamz)
- "I cant stand when it gets things wrong on the most basic of stuff." (j3k3)
4. 从创作到审查的角色转变
- 观点:LLM将开发者从创作者变为审查者,而审查比创作更耗费精力。
- 引用:
- "LLMs fundamentally shift work from generation to verification... reviewing is harder work per unit of output than writing." (stainlu)
- "Reviewing this junk has become exhausting... I don’t want to read your code if you haven’t bothered to read it yourselves." (rednafi)
5. 技术与社区的负面影响
- 观点:LLM技术本身有趣,但过度炒作和社区氛围令人疲惫。
- 引用:
- "llms aren’t exhausting it’s the hype and all the people around it." (anthonySs)
- "It’s like we are gleefully running to the 'Laughterhouse'..." (dinkumthinkum)
6. 异步与适度使用的建议
- 观点:异步使用LLM并控制任务数量可减轻负担。
- 引用:
- "I find working more asynchronous with the agents help... I work across a few different tasks at my own pace." (olejorgenb)
- "It’s ok to let them be idle while you finish up what your doing." (olejorgenb)
7. 结构性问题与改进空间
- 观点:LLM的局限性是结构性的,但改进和抱怨都是合理的。
- 引用:
- "Things can be improved. You are allowed to complain about anything." (sigbottle)
- "The exhaustion isn’t primarily from bad prompts... it’s that the LLM turned you from a writer into a full-time reviewer." (stainlu)