早上听了 Walker 采访 世界著名 AI 专家、麦肯锡数字化高级合伙人 Rodney Zemmel 谈到了他的《华尔街日报》最畅销新书 The McKinsey Guide to Outcompeting in the Age of Digital and AI，一些要点： 1.人工智能🌊有很多好处，但也有一些缺点：70%的资源被浪费和浪费。由于培训材料存在偏见，人工智能存在偏见。 AI 幻觉/疯狂想象 但最令人担忧的是 AI Fake，几乎无法辨别差异。 所以未来除了设置护栏/法律之外，人们的信任也是建立在消息来源的基础上 2. 之前，人工智能是更具体的，即专家系统类型，但是从去年左右开始，法学硕士启用了生成式人工智能，一种模型适用于不同的现实世界场景 3. AI 火车🚂还处于非常早期的阶段，做生意的方式正在被 AI 改变，刚刚开始，最好上车，否则有在车站掉队的风险 4. 必须是公司高层自上而下的驱动，制定目标并制定可实现目标的路线图，但根据麦肯锡的数据，只有大约 1/3 的人获得了经济回报，结果没有，许多公司在技术上投入过多但结果微乎其微 5. 只有大公司才有资源建立自己的 LLM，更多的公司会使用现成的，然后修改以满足自己的需求 6. 数据发现，由于共同的培训材料，许多 LLM 模型收敛/行为相似 7. 公司路线图应分为 3-6 个月/2-5 个较小的项目块，并具有现实的可交付成果，以检查每个小里程碑的结果，而不是 2-3 年的宏伟计划…… 8. 许多传统公司也在改变他们的招聘策略：不再雇佣 IT 主要用于维护，现在寻找具有真正有价值的职业道路、内部成长和领导机会的技术人才（听说这里，给 CS 但毕业就业市场鸭梨山大的）童鞋们一点建议，不要光👀着高科技公司，传统行业也需要人才） 9、麦肯锡数字化搭建的新数据库，基本整合了麦肯锡内部的 42 个数据库，利用 GenerativeAI 梳理了麦肯锡所有的用户案例，对于客户面临的任何问题，可以提供 4-5 个麦肯锡内部专家和最佳实践，可以由任何麦肯锡员工拉动 10.人工智能目前处于早期阶段，主要用于自动化钻孔任务，以提高效率。在商业中采用人工智能，是一个旅程/过程，而不是一夜之间的奇迹，还很早，随时加油 

早上听了 Walker 采访 World renowned AI expert, McKinsey Digital senior partner Rodney Zemmel, talked about his WSJ best selling new book, some key points:

1. Lots of benefits of AI 🌊, here are some downsides: 70% of resources misspent and wasted. AI-biased due to training materials being biased. AI hallucination/crazy imagination But the most worrisome is AI Fake, almost impossible to tell the differences. So in future, aside from setting guardrails/laws, also people’s trust is based on the sources of the news
2. Before, AI was more specific ie expert system type, however since last year or so LLMs enable Generative AI, one model apply to different real world scenarios
3. AI train 🚂 is still on very early stage, the way to do business is being transformed by AI, just getting started, better hoop on or risk of being left behind in station
4. Must be top down driven from the top of the company, setup goal and have roadmap with achievable objectives, but based on the McKinsey data, only about 1/3 is getting financially reward as a result not, many companies over spent on technology but marginal results
5. Only large companies have resources to build their own LLMs, many more will use off the shelf and then modify to suit their own needs
6. Data found many LLM models converge/behave similarly, due to common training materials
7. Companies roadmaps should break into 3-6 months/2-5 smaller projects chunks, with realistic deliverables, to check on result at each small milestone, not 2-3 year huge grand plan…
8. Many traditional companies also changing their recruiting strategies: instead of hiring ITs mainly for maintenance, now looking for tech talents with real rewarding career path with internal growth and leadership opportunities (听到这里，给 CS 毕业但就业市场鸭梨山大的童鞋们一点建议，不要光👀着高科技公司，传统行业也需要人才）
9. A new database built by McKinsey Digital, basically consolidate 42 internal databases at McKinsey, use GenerativeAI to sort out all user cases In McKinsey, for any problem faced by clients, can provide 4-5 internal McKinsey experts and best practices, can be pulled by any McKinsey associates
10. AI at this early stage, mainly for automated boring tasks to improve efficiency. Adopting AI on business, is a journey/process and not an overnight miracle, still very early, hoop on anytime