Our Workplace LLM Mass Delusion

我们职场中的大语言模型集体妄想

10 Jun, 2026 2026年6月10日

I can’t help but wonder whether we will look back on this AI hype in the workplace with confusion and embarrassment. If we indeed progress into a future where the bubble will burst, models will further close up, become too expensive for the average user, enshittified, or really specialized for specific fields and most promises end up not fulfilled, how will employers everywhere play this off? How will employees recover from witnessing this cultish environment suddenly dropping off as if nothing happened? 我不禁在想，未来当我们回首这段职场 AI 狂热时，是否会感到困惑和尴尬。如果未来泡沫真的破裂，模型变得更加封闭、对普通用户而言过于昂贵、质量恶化（enshittified），或者变得过于专业化而导致大多数承诺无法兑现，各地的雇主将如何收场？当员工们目睹这种狂热的氛围突然消散，仿佛一切从未发生过时，他们又该如何恢复？

My employer, for example, struggles with funding. Open positions are not to be filled and will just fall away; employee bonuses for great work have been permanently cancelled 2 years ago due to the tense financial situation; necessities fell away with a message to just “find a way to deal with it”. Several departments are completely overworked with no help in sight, and are just asked to cut corners. Important licenses and databases are dropped to save money. This is the backdrop to our AI adoption in the workplace. 以我的雇主为例，公司正面临资金困境。空缺职位不再招聘，直接被裁撤；由于财务状况紧张，员工的绩效奖金在两年前就被永久取消了；必需品被削减，只留下一句“自己想办法解决”。多个部门工作量严重超负荷，却得不到任何援助，只能被要求“走捷径”。为了省钱，重要的软件许可和数据库被停用。这就是我们职场采用 AI 的背景。

Still, somehow, there is enough money to hire consultants that advise to go all in on AI for a possible future where money can be saved, and enough money to pay external companies for LLM workshops and seminars for employees for years, and enough money to pay for licenses of both ChatGPT and Copilot. That means: The employee bonuses that should go to all the hardworking employees, and the money to further support our work, is going to grifters, security risks, bad workshops that are not teaching anything remotely usable for our work, and technofascists. 然而，不知为何，公司却有足够的钱聘请顾问，建议我们全力投入 AI 以期在未来节省开支；有足够的钱支付外部公司多年来的大语言模型（LLM）研讨会和讲座费用；也有足够的钱支付 ChatGPT 和 Copilot 的许可费。这意味着：本应发给辛勤工作的员工的奖金，以及本应支持我们工作的资金，全都流向了骗子、安全隐患、毫无实用价值的糟糕研讨会，以及技术法西斯主义者手中。

Not only that! We have recurring house-wide meetings where groups are asked to show off their LLM projects. They register them, try them out for a couple months, and then come back presenting their results. I have attended all of these meetings so far, and there was not a single one that actually worked out. All projects ended with the conclusion that this isn’t workable, that this isn’t saving time, or that it over-complicates things. Hundreds of people, different teams, people enthusiastic about AI, all kinds of projects, and there wasn’t a single success. 不仅如此！我们还会定期召开全公司会议，要求各个小组展示他们的 LLM 项目。他们注册项目、试用几个月，然后回来展示结果。到目前为止，我参加了所有这些会议，没有一个项目真正成功。所有项目最终的结论都是：这行不通，这不能节省时间，或者这只会让事情变得更复杂。数百人、不同的团队、对 AI 充满热情的人们，尝试了各种各样的项目，却没有一个成功的案例。

All kinds of workshops, “prompt-engineering”, custom GPTs and skills, pre-prepared documents and templates could not make something truly effective and reproducible in our field of work (not anything coding related!). It was a messy gamble every time. It took a significant amount of time to fine-tune everything, to repeat the task, to verify the output, and correct mistakes before continuing with the rest of the workflow. Not considering this or that document, hallucinations, inability to fill in documents correctly or edit them were the biggest complaints. Even on an Enterprise license, the restrictions were too great. 各种研讨会、“提示词工程”、自定义 GPT 和技能、预设文档和模板，都无法在我们（非编程相关）的工作领域中实现真正有效且可复制的成果。每一次尝试都是一场混乱的赌博。我们需要花费大量时间去微调、重复任务、验证输出并纠正错误，才能继续后续的工作流程。忽略某些文档、幻觉问题、无法正确填写或编辑文档，是大家抱怨最多的地方。即使是在企业版许可下，限制也太多了。

But wait, there’s more! We also have house-wide meetings where employees show off how ChatGPT can be used regardless of specific projects; just general use cases for the workday. Let me tell you what great things were shown off. For one, it was shown that you can ask the bot how it feels today. That wasn’t presented as a joke, or being sarcastic; no, it was shown very seriously, I guess under the guise of how cool and futuristic and human it is. 还没完！我们还有全公司会议，员工们展示 ChatGPT 如何在具体项目之外使用，即日常工作中的通用案例。让我告诉你他们展示了什么“伟大”的成果。其中一个案例是展示你可以问机器人今天感觉如何。这并不是作为笑话或讽刺提出的；不，它是非常严肃地展示出来的，我想大概是打着“它有多酷、多前卫、多像人”的幌子。

I’m getting really upset here at the point of writing this, because I have to fight hard to get the funding for the database my team needs for my work and have to justify it every year, and I know that in any other contexts, or just 5 years ago, they would have laughed in your face if you suggested to get a subscription in the thousands to enable employees to have a pointless conversation with a bot. Hello, we have shit to do over here, departments are drowning in work, and you wanna have software that talks to you? That would have been the response, and it is the correct response still!! People like that need to be treated like the fools they are, and we need to challenge them more! 写到这里我真的非常生气，因为我必须拼命争取团队工作所需的数据库资金，并且每年都要为此辩解。我知道在任何其他背景下，或者仅仅在 5 年前，如果你建议花几千美元订阅费让员工与机器人进行毫无意义的对话，别人一定会嘲笑你。喂，我们这里还有一堆烂摊子要处理，各个部门都忙得不可开交，你却想要一个能跟你聊天的软件？这本该是（也依然是）最正确的反应！！这种人就该被当成傻瓜对待，我们需要更多地质疑他们！

Next up was the great use case of downloading the cafeteria menu (which is a 1 page nicely designed Excel sheet, like a timetable, showing the different options for each day) from the intranet, giving it to ChatGPT, and asking it what’s for lunch on Wednesday. I wish I was joking. I WISH! The bot spat out a longer answer than reading the entire sheet would be. Downloading and uploading and writing the prompt took longer than just reading the sheet. You can see what’s for lunch on Wednesday with one glance already. No bot needed! 接下来的“伟大”案例是从内网下载食堂菜单（那是一张设计精美的单页 Excel 表格，像时间表一样显示每天的菜品），喂给 ChatGPT，然后问它周三午餐吃什么。我真希望我是在开玩笑。我真的希望！机器人吐出的答案比直接阅读整张表格还要长。下载、上传和编写提示词所花的时间，比直接看表格还要久。你一眼就能看出周三午餐吃什么，根本不需要机器人！

The other general use case presented to us (by our head of IT, no less) was that if we are not sure whether something is a spam mail, phishing attempt, a mail with a suspicious attachment, whatever… we should save it to our Desktop, upload it to ChatGPT and ask it. Good god. I am still in disbelief. I’m sorry, but I don’t want the less technically inclined employees among us to save anything shady onto their work laptop. Come on now. What is happening? Have we lost our minds? 另一个（由我们的 IT 主管亲自）展示的通用案例是：如果我们不确定某封邮件是否为垃圾邮件、钓鱼攻击、带有可疑附件的邮件等等……我们应该把它保存到桌面，上传给 ChatGPT 并询问它。天哪。我至今仍感到难以置信。抱歉，我可不想让我们中那些技术能力较弱的员工把任何可疑的东西保存到工作电脑上。拜托了，这到底是怎么了？我们都疯了吗？

Intentional or not, AI is seemingly great at amplifying the Dunning-Kruger-Effect in people, making everything they attempt with it seem smarter and justified to them, packaging every fart in a nice bow that makes it seem deep initially. People can pretend they’re now doing something really important and groundbreaking while using the tool for completely mundane and worthless tasks that are better handled differently. 无论是有意还是无意，AI 似乎非常擅长放大人们的“达克效应”（Dunning-Kruger Effect），让他们觉得用 AI 尝试的每件事都显得更聪明、更合理，把每一个屁都包装上漂亮的蝴蝶结，让它起初看起来很有深度。人们可以假装自己正在做一些非常重要且开创性的工作，而实际上却是在用这个工具处理完全平庸且毫无价值的任务，而这些任务本可以用更好的方式解决。

Defenders of the tech can feel like they’re part of something big and revolutionary and fantasize about the day they will be proven right and all the critics will shut up or apologize (like my conspiracy theorist dad, who still clings to the same prophecies after over a decade, hoping to be ahead of the curve and right for once in his disappointing life). It’s sad, because it feels like a completely out-of-control delusion; you see smart and capable people with lots of responsibility at work suddenly turn into a shill for these AI companies without any rhyme or reason. 这项技术的拥护者们会觉得自己参与了一场伟大的革命，并幻想有一天他们会被证明是正确的，所有的批评者都会闭嘴或道歉（就像我那个阴谋论者父亲，十多年来一直坚持同样的预言，希望自己能走在时代前沿，在他那令人失望的人生中至少赢一次）。这很可悲，因为它感觉像是一种完全失控的妄想；你看到那些聪明、有能力且肩负重任的职场人士，突然毫无缘由地变成了这些 AI 公司的托儿。