Summary of the Key Points
This article reveals a harsh new reality in the age of AI: AI models are replacing humans by “distilling” human knowledge, particularly those intangible skills that are difficult to articulate. This has created a “recursive distillation food chain” where each level—individuals, companies, and industries—absorbs knowledge from the level above before discarding it. Although Microsoft CEO Satya Nadella is aware of this mechanism, he is forced to participate in it (warning about the loss of knowledge while simultaneously laying off employees). The “moat” that once protected experts’ intellectual assets is being drained by AI. Layoffs are no longer a product of economic cycles but a routine part of the AI machine’s operation. Ordinary people must protect their own learning processes to prevent their knowledge from becoming fuel for others.
1. The Waste of Expert Knowledge
You may have heard that AI needs data for training, but now it is trained using real experts rather than just cat pictures. For example, the American company Mercor hired 30,000 contractors—doctors who study general relativity, voice actors who speak Hebrew, and doctors from Rwanda—for a salary of up to $225 per hour. Their job was simply to transfer their professional knowledge to the AI models. Once the models learn enough, these experts become obsolete and are discarded, just like the residue after brewing traditional Chinese medicine.
Another company, Handshake, went even further with a “banker benchmark test” where real people acted as clients and investment bankers, replicating every interaction within Goldman Sachs down to the minute details, using 90 evaluation criteria. The goal was to understand what Goldman Sachs employees do, so that AI could learn those skills. A New York Times reporter added a poignant comment: “There’s no need to guess what will happen to Goldman Sachs’ employees next.”
Some of these experts worked for 72 consecutive hours, surviving on short naps on sofas, but they could not escape the fate of losing their jobs once the models learned enough.
2. Nadella’s Paradox
Satya Nadella is a discerning individual who has published two lengthy articles warning about how AI is gradually distilling human knowledge. He mentions the “reverse information paradox”: in the past, those who sold information feared giving it away for free; now, buyers of AI services fear even more that to get useful models, they must feed their proprietary knowledge into them—paying twice (once with money and once with knowledge).
Ironically, in the same week, Nadella announced the layoff of 4,800 employees at Microsoft. The reason? Nadella himself is part of this food chain. Microsoft is concerned that OpenAI (a company it invests in) could drain its intellectual assets (for example, by learning from Microsoft’s software and workflows on its cloud platform). Therefore, layoffs are necessary to optimize the company’s structure; otherwise, it could be overtaken by OpenAI.
This is not hypocrisy but a form of “structural collusion.” No matter how clear you see the situation, you have to adapt to the machine’s logic: if you don’t “distill” others’ knowledge, someone else will do it for you; if you don’t lay off employees, the company won’t survive. Nadella is both a “prophet” (who sees the mechanism) and an “accomplice” (who participates in this process), as there is no pause button in this endless cycle.
3. The Recursive Distillation Food Chain
This distillation is not isolated; it’s like a Russian matryoshka doll, where each layer repeats the same process:
- Individual level: Experts are “distilled” by AI laboratories (such as Mercor’s contractors).
- Corporate level: Microsoft is “distilled” by OpenAI, which learns from Microsoft’s knowledge.
- Industry level: Advanced AI models are “distilled” by cheaper models (for example, free and open-source models that devalue larger, more expensive ones).
The rule of this chain is not “the big fish eats the small fish” but “whoever controls the learning process has the upper hand.” Whichever model absorbs your knowledge gains an advantage over you. For instance, although Microsoft is larger than OpenAI, OpenAI can still affect it because Microsoft’s knowledge goes into its models.
4. The Disappearance of Hidden Knowledge Moats
Experts used to feel secure because of their “hidden” skills—those abilities that couldn’t be written down in resumes, such as intuition or the ability to spot issues with reports at a glance. It took years for experienced teachers to pass on their knowledge to apprentices because these skills were impossible to codify.
AI has changed all this. It doesn’t need you to articulate your skills; it can extract them from your actions and daily work patterns. This effectively drains the moats that protected experts’ knowledge. Moreover, AI continues to improve on its own: what used to require manual processes (like training AI models) is now automated, making even the jobs related to training obsolete.
5. A Survival Guide for Ordinary People
In the age of AI, tasks that involve mere execution (such as basic investment analysis or standardized customer service) will likely become less valuable. However, skills that cannot be easily distilled—such as judgment, interpersonal skills, creativity, and cross-domain pattern recognition—will become increasingly important. Ordinary people should do two things:
1. Set clear boundaries for trust: Keep the things that AI cannot replace—decisions about what information to trust, insights into hidden costs, personal taste, and connections. Don’t hand over all of your cognitive abilities to AI.
2. Build your own learning processes: If you invest an hour of work and let your knowledge flow into someone else’s model (like working for Mercor), it’s like wasting it; if you integrate it into your own system (by writing articles or gaining experience through projects), it becomes a valuable asset that can generate long-term benefits.
Nadella once said, “You can outsource tasks and even entire jobs, but never your learning.” This advice is meant for CEOs, but it’s especially important for ordinary people to remember: don’t become empty containers; instead, be the one who holds the “recipe” to your own knowledge. Keep your learning process under your control.
This distillation machine will continue to operate, but you can decide where your knowledge goes. Don’t let yourself become just another part of its cycle—don’t end up being discarded like waste.