OpenAI 正在加速吸纳 Thinking Machines Lab 的研究人员。本周三，OpenAI 应用平台主管宣布重新聘用该实验室的两位联合创始人，他们曾于 2024 年底离职。知情人士称，Thinking Machines 内部曾指控其中一人涉及严重不当行为，并在同日将其解雇；而 OpenAI 表示相关招聘已筹划数周，且不认同其伦理指控。除两名联合创始人外，另一名前员工也将回归，且未来数周预计至少还有两人加入。这一事件被视为延续了 2023 年 OpenAI 短暂罢免 CEO 的内部动荡。

此类人事争夺反映了 AI 行业持续的权力重组。多家头部实验室近年均出现联合创始人或核心研究人员出走。尽管争议频发，研究人员的流动性仍受到巨额资金支持，知情者称在当前环境下，数十亿美元规模的种子轮融资仍可迅速完成，使行业震荡呈常态化。

与此同时，AI 实验室正系统性推进“代理”自动化以替代知识型工作。过去一年，数据供应商开始向顶尖专业人士支付每小时 100 美元以上，以收集真实工作样本，用于训练 AI 在法律、医疗、咨询和金融等领域的能力。这些数据被用来构建模拟“环境”，让模型反复练习企业软件操作。尽管成效仍待验证，但与一年前相比，AI 代理在稳定性和复杂任务处理上的提升已十分显著。

OpenAI is accelerating its recruitment from Thinking Machines Lab. On Wednesday, OpenAI’s head of applications announced the rehire of two cofounders who had left in late 2024. A source said Thinking Machines alleged serious misconduct by one cofounder and fired him the same day, while OpenAI stated the hires had been planned for weeks and rejected ethical concerns. Another former employee is also rejoining, with at least two more expected in coming weeks. The episode echoes internal turmoil dating back to OpenAI’s brief 2023 leadership crisis.

The episode reflects ongoing power reshuffling across AI labs. Several leading companies have seen cofounders and senior researchers depart in recent years. Despite persistent controversy, researcher mobility remains high, supported by abundant capital; sources say billion-dollar seed rounds can still be raised quickly, making disruption a structural feature of the sector.

At the same time, AI labs are intensifying efforts to automate knowledge work through agents. Over the past year, data suppliers have paid more than $100 an hour to top professionals to provide real work examples. These datasets train models in fields such as law, health care, consulting, and banking, often via simulated “environments” that mimic enterprise software. While reliability is still unproven, agents have advanced markedly over the last year, suggesting broader workplace automation is approaching faster than before.