<div id="readability-page-1">Is this a deep learning neural network, with blue inputs, green hidden layers, and a red output? Or is it a knocked-over org chart, with blue entry level jobs, green middle managers, and red executives? Yes. Yes it is. It is both. getty Is there any room for humans in the post-AI world? Or will the job market itself fade away to become a legend of what once was? Perhaps it depends how senior you are.  If you knock an org chart over on its side, the layers of a typical corporation look very similar to that of a deep learning neural network. At the far left are the entry-level positions, at the far right are the senior executives, and the middle holds all the, well, middle managers. For a neural net, the far left is the input, the far right is the output, and the middle layers act to increasingly abstract and generalize the earlier layers.  For some time, it has seemed as if entry level jobs were hurting the most due to AI. It’s not that entry-level employees were being fired; they simply weren’t being hired as much. According to the Bureau of Labor Statistics, the unemployment rate for recent college grads has now spiked to 9.7%. This is the highest it has been in four years. Excluding summer 2021 and Covid, this is the highest it has been in over a decade.  With AI, the story makes sense, at least on the surface. Why hire humans to do data entry, update presentations, or summarize documents when it can be done thousands of times cheaper and faster with AI? If the main value of entry-level jobs is menial tasks like this, then they will join the list of other antiquated jobs like ticker tape clerks, telegraph operators, and typists.  But the story falls apart quickly, and the network graph above helps explain why. The resemblance between the org chart and the neural net is deeper than it looks. The entry-level jobs really are inputs to the firm. Did you think the main reason companies hired entry-level workers was to do entry-level work? That would be too on-the-nose. The real reason may be much more subtle.  Think of the history. Think of all the entry-level jobs that have disappeared, and yet the entry level itself has remained. Ticker tape operators got a sense of how markets behaved, absorbing the rhythm of the trading floor long before they handled a single order themselves. Telegraph operators learned the pulse of information flow. Typists represented an informal collection of institutional memory.  MORE FOR YOU Yesterday’s data entry jobs gave recent graduates immediate exposure to sensitive and important information. Updating presentations helped them learn important communication and persuasion tips. Summarizing documents meant you had to learn things quickly and relatively deeply.  In all cases, apprenticeship by osmosis brought inexperienced people right into the center of the action. It built trust, familiarity, and reliability in an environment that transferred tacit and implicit knowledge. Think of the sales assistants and trader assistants on a traditional open-plan trading floor. Is there any better learning environment?  Perhaps the main value of entry-level jobs is not and has not really been in the menial tasks. Perhaps their main value has been as an investment for the future. Just like a neural net, they are an input to the firm. These jobs propagated context.  The particular jobs that are at the entry level can change, but the core concept will likely remain, because firms do need an input layer of context, a way to transfer and accumulate knowledge. They need the entry-level hires to grow, to eventually replace the aging middle layers.  The middle layers have seen better days. Unlike the entry level layer, where hiring has slowed, the middle layers are being actively deleted, at firms of all sizes.  In small and medium sized firms, according to a July 2025 report from Gusto, layoffs for middle managers have surged three-fold since 2022, and hiring has collapsed by half.  Among large firms, a not uncommon example is Amazon’s recent move to lay off thousands of middle managers, explicitly citing AI.  Like the reduced openings in entry-level jobs, the pruning of middle managers is a widespread, economy-wide phenomenon.  One of the most mysterious aspects of neural nets is the power of dropout layers. Almost bizarrely, introducing layers simply to forget about half of what the previous layer learned can help enormously. It helps to generalize and avoid overfitting. If something is really, really important in an earlier layer, then it should be able to burst through even after dropping half of the middle managers. Perhaps this is part of what is happening with the economy now.  The final layer, the executives, are the output layer. That is where all the preceding complexity collapses into a decision: buy or sell? Invest or wait? Expand or contract? Stay the course or pivot?  In a corporation, the executives determine whether the firm is aligned with reality. They represent the loss function. If their earlier forecasts were wrong, they need to backpropagate changes to the weights, mainly in the hidden middle layers, to get the company back on track.  Perhaps some day, truly AI-native firms like Elon Musk’s forthcoming Macrohard may even replace executives in this final decision making. For now, however, the loss function appears to be trading at a premium.  According to an analysis last month by Gallagher, median CEO pay climbed 9.7% in 2024. This was nearly double the growth rate of the previous year and, most importantly for our story here, it was not driven by larger companies, but by smaller ones. Firms with less than $50 million in revenue saw their CEO pay surge by nearly 45%!  What is going on here? New entry-level jobs are becoming scarcer. Half of middle managers are getting actively laid off. And executives are earning more. The answer may again go back to the neural net analogy.  What we may be witnessing is the rise of the context monster.  Whether with agents or workflows or one-shot interactions, AIs can do human work exceptionally well if they are provided with sufficient context. The skills that appear now to be most in demand are therefore not the tasks themselves anymore, but the ability to communicate all of the relevant context to the AI in an efficient way.  Good executives have a lot of the context. Great executives perhaps have all of it. They can explain what every role does and why. Such executives become more valuable.  Some middle managers are excellent at context as well, while others either lack sufficient context themselves or have difficulty articulating it clearly to a machine. That may explain the wide variability in those middle layers.  For the entry-level jobs, there can be no neural network without an input layer. Firms need fresh eyes and new perspectives. But they do not need the menial tasks that people think entry-level jobs are for.  What does that mean for students and young people looking to land a good position in the job market?  Probably it is less valuable to focus on skills now that AI can perform the same tasks better. Your edge is no longer in the task. It is in the translation. Don’t build a portfolio of skills. Build a portfolio of workflows. Show that you can be the bridge between the raw data and the final decision, across a variety of industries. Show that you, too, are a context monster.</div>

<p>這篇文章將企業結構與神經網絡進行比較，強調了人工智能對職位角色的影響。入門級職位類似於輸入，面臨招聘下降，最近畢業生的失業率為 9.7%。中層管理者被比作 dropout 層，經歷裁員，而高管則蓬勃發展。儘管人工智能的影響，入門級角色對於知識轉移和未來增長至關重要</p>

上下文怪物與丟棄層：就業市場如同神經網絡