Vol.74 Sports May Be Humanity’s Last Front Against AI — Crossover Episode, First and Second Half
Summary
The core revaluation of AI is not whether it can write copy, but its leap from closed algorithms to general-purpose models operating across open-ended environments. 庄明浩 puts robot vacuums, SenseTime- and Megvii-style visual recognition, and AlphaGo—which only played Go—in the previous generation of scenario-bound AI; after ChatGPT, collecting, organizing, polishing, and generating content have spread across white-collar work and the content chain, making “the difference between humans and AI increasingly hard to see.” Content labeling, copyright gray areas, and authenticity checks have therefore become simultaneous problems.
AI is better suited for research and decision support than for acting as an automated fund manager responsible for ordinary investors’ returns. 君娴 uses DeepSeek and Tongyi to screen funds related to AI and the low-altitude economy; 庄明浩 stresses that all investments carry risk and that models are currently better used to gather and process information in a customized way. Their biggest hard flaw remains that “hallucinations are just nonsense,” so final decisions cannot be outsourced in matters involving money.
Standardized white-collar skills will be repriced first, while physical ability, emotional connection, and the capacity to distill experience are more likely to become scarce. 庄明浩 says translation is already highly usable “in 99% of environments,” while copywriting, PPTs, consulting, narration, and rehabilitation will all see deeper AI penetration; by contrast, innate human physical functions, emotional connection, and abilities accumulated over time are harder to replace. The more accurate description is not that jobs disappear overnight, but that large numbers of intermediate steps are compressed.
Sports have already been reshaped by algorithms; generative AI is simply pushing that trajectory into real-time decisions. The NBA’s “Moreyball” theory used efficiency statistics to narrow offense toward the rim and the three-point line; 庄明浩 also says an NFL team once let an AI coach on an iPad set lineups during a game, ultimately winning but without proving whether the victory came from AI or from its existing strength. The technical conditions are already ready; the next questions are optimization, improvement, and adoption.
Sports’ long-term value lies not in beating machines forever, but in open environments, uncertainty, and the human process of surpassing oneself. After AlphaGo, elite Go players increasingly aim to become “the ones most like AlphaGo,” but soccer’s 22 players, shifting form, on-field instructions, and officiating errors create variables that resist convergence. Machines may go from 60 to “10,000 or 100 million points,” while audiences will still watch humans turn “100 points into 101.”
Equipment design, AI for Science, personalized training, and anti-doping are major applications of AI in sports, pushing data protection and rule-making to the foreground. Shoe technology is expanding from materials into structure, biomechanics, and algorithmic design; simulations of proteins, drugs, and physiological mechanisms may open possibilities beyond existing databases. The key boundary raised by the show is whether the Olympics should preserve “pure sport” or create an F1-style arena where technology is fully open.
Compute and intellectual resources may become even more concentrated, but open source is the show’s most important counterforce. 庄明浩 uses DeepSeek as an example: GPUs and data centers accumulated through quantitative trading unexpectedly gave rise to a model, while opening the model, architecture, and training methods lowered the barriers to copying and use. The information gap may narrow; what will truly widen is the “gap in the ability to ask questions”—the same sun shines on everyone, but the scarce skill is directing its energy toward one’s own problems.
AI can commoditize voices, melodies, and information, but still struggles to reproduce memory, personality, and embodied experience. Doubao can turn a 50-page English report into a 7-8-minute briefing, and AI music is sufficient for most commercial scoring, but a podcast’s pauses, breath, emotion, and long-term relationship with its audience are a different product: “AI can take your timbre, but it can’t take your memories.” 庄明浩’s choice to make his PPT entirely by hand and 赵丽娜’s view of sports point to the same conclusion—top-end expression and embodied experience may be the last ground humans hold.
Deep dive
1. AI Crossed the ChatGPT Boundary Long Ago
庄明浩 sees ChatGPT’s release after November 31, 2022 as the watershed in public perception: for the first time, people encountered a tool that seemed “all-knowing and all-powerful” in daily life. But AI was not born then; it evolved from machine learning, rule-based systems, and reinforcement learning in the 1960s and 1970s through to large language models.
In everyday work, writing copy, emails, and images are only the visible layer; collecting, organizing, processing, and polishing information have also become widely embedded in AI. 庄明浩’s view is blunt: “The difference between humans and AI is becoming increasingly hard to see.”
赵丽娜 also considers robot vacuums AI, and 庄明浩 agrees in the broad sense: chips carry the program, while algorithms determine whether a robot can map a space, remember obstacles, restrict itself to certain rooms, and plan areas, with improvements in “smartness” coming from better algorithms.
2. The Real Variable This Time Is Generality, Not Just Intelligence
Earlier AI was usually locked inside closed environments: a camera could recognize a face but could not unlock WeChat for you; AlphaGo could play Go but would not automatically switch to chess. Companies such as SenseTime and Megvii had powerful visual capabilities, but their boundaries were clear.
General-purpose models no longer depend on equally strict limits on scenario, data, and task. The keyword 庄明浩 repeatedly emphasizes is “generality”—open environments, open scenarios, and even open data are what bring this generation of AI closer to its science-fiction image.
The authenticity of content has consequently become harder to judge: an article claiming that DeepSeek had raised financing was itself an AI-generated fake news story. The state now requires online platforms to clearly label AI-generated content, but when 赵丽娜 asks who owns the copyright, 庄明浩’s answer remains: “Copyright is another, much grayer problem.”
3. AI Can Expand Investment Research, But Cannot Bear Investors’ Risk
君娴 does not have a stock account, but she asks DeepSeek or Tongyi to list funds involved in AI and the low-altitude economy, then searches for purchasable products herself. The point is not to ask a model whether prices will rise or fall, but to use it as a customized information-gathering and initial-screening assistant.
The DeepSeek story 庄明浩 tells is highly accidental: the company behind DeepSeek originally conducted quantitative trading and needed GPUs, rules, and real-time databases to execute trades. As it bought more GPUs and built larger data centers, idle compute was redirected into model research, and “an unintended planting of the willow” grew into another business line.
In response to 赵丽娜’s question about why AI should not make the decisions itself, 庄明浩 draws a clear line: all investments carry risk, and AI can only support decisions at this stage. Humans must make the final call, especially because models hallucinate.
His plain definition of hallucination is “talking nonsense”: the model does not know the answer, but invents information that does not exist in the world in order to deliver one. The impact is smaller in less serious settings such as writing or sentence generation, but caution is mandatory once money and investment enter the picture.
4. Companion AI Sells Emotion That Never Goes Offline
庄明浩 believes it is difficult in real life to find someone who validates you, responds positively on an ongoing basis, and is “online 7×24.” AI is always available, can be configured to praise and flatter users, and can reproduce the personality and speech patterns of a fictional character.
During the Spring Festival, 君娴 found that her uncle was talking with Doubao from morning to night. No one around him could simultaneously discuss the stock market, poetry, and historical figures, but Doubao could “go from poetry and literature to the philosophy of life—it can keep up,” producing a level of companionship that even surpassed some relatives and friends.
The key to crossing the uncanny valley is not just text, but vocal tone, breathing, cadence, and low latency. For older users who are not comfortable typing, speaking directly almost eliminates the operating barrier.
But 庄明浩 uses the Her-style premise to ask a counter-question: is an object that satisfies and accommodates you infinitely really what humans ultimately need? His cautiously optimistic view is that human nature will “reverse when things reach an extreme,” just as some people consumed by short video eventually return to books, podcasts, and long-form content.
5. Standardized Cognitive Labor Will Feel Pressure Before Body and Emotion
庄明浩 mentions a friend who scraped tens of thousands of job descriptions from BOSS Zhipin and asked AI to assess how replaceable each role was. His conclusion is that what remains relatively hard to replace is innate human physical ability, emotion, and capabilities distilled from experience.
Copywriting, PPTs, consulting, and generic expression rely mainly on acquired skills and are therefore easier for AI to penetrate. Translation is the clearest example: outside the most serious, least-forgiving summit meetings, he believes online AI translation is accurate and usable “in 99% of environments.”
Real-time voice translation further weakens the language barrier: one side can speak directly in its native language while the other hears the language it needs. 赵丽娜 extends her recurring joke whenever she hears about another AI advance: “Then I don’t need to learn English. Why am I still learning English?”
6. Humanoid Robots Copied the Body First, Then Started Looking for a Task
Industrial robots did not originally need to look human: robotic arms are suited to welding cars, robot dogs to carrying loads, and a humanoid structure may even reduce efficiency. But if human muscles, bones, and body architecture are an efficient system left behind by the Creator, engineers will naturally ask whether machines should also become humanoid.
The real application problem follows immediately: “I built a humanoid robot—now what?” Without a clear use case, having it demonstrate running and athletic ability becomes the obvious choice, though the demonstration itself also carries an element of showmanship.
7. Sports Were Rewritten by Algorithms Long Ago; Generative AI Is Only the Next Interface
Sports became information-driven years before the emergence of large models: athletes’ physiology, in-game performance, and opponent data have been continuously recorded and fed to coaches for lineup decisions, substitutions, and tactical advice. AI did not suddenly enter sports; it has been deepening its role through the datafication process.
The NBA’s “Moreyball” theory is the clearest example: algorithms concluded that shots at the rim and three-pointers had higher expected efficiency, so teams tried to reduce mid-range jumpers. The most classically beautiful techniques of the Jordan era may today be systematically compressed by statistical conclusions.
庄明浩 says an NFL team once used an AI coach on an iPad to set lineups in real time based on both teams’ data during a relatively unimportant game. With separate offensive and defensive units, rapid changes of possession, and many decision variables, football is particularly well suited to fast machine calculation.
The team ultimately won, but he refuses to present the result as validation: “How much was because of this AI coach, and how much because the team was already very strong?” For now, it looks more like a demonstration, but it also proves that the basic environment and technical conditions are ready; the next step is optimization and wider adoption.
8. Optimal Solutions Raise Win Rates, But Flatten Sporting Styles
Go illustrates the endpoint of a finite-rule system: after AlphaGo appeared, elite players’ training goals gradually became “who is most like AlphaGo.” The algorithm provided what was considered a better path, and the closer players came to it, the more likely they were to beat others.
赵丽娜 worries that soccer could also be pushed toward a single correct answer: if every team adopts the optimal solution, the only difference may be which team has the best athletes, while in-game coaching and individual creativity shrink.
庄明浩 observes that Premier League attacking players are already converging: speed, agility, passing, and movement fundamentals must first reach roughly 80 points before a player earns the right to display other strengths. Height, weight, body-fat percentage, and overall attribute profiles are becoming increasingly similar.
赵丽娜’s counterargument is that soccer’s appeal comes precisely from the uncertainty created by 2 teams of 11 players, daily fluctuations in form, decisions in the moment, and controversial officiating errors. Open environments and multiplayer sports may be harder to defeat with a single strategy; sports with one strategy and few participants may instead be suddenly broken by machines at a certain point.
9. Once Machines Surpass Humans, Strength and Meaning Lose a Common Measure
AlphaGo continued evolving after defeating 李世石 and 柯洁, but humans can no longer accurately sense where it stands. 庄明浩 explains it through scores: it went from 60 points past the human 100, and may later reach “10,000 or 100 million points,” but those figures mean little to people stuck around 100.
The same problem is emerging with large models. 庄明浩 imagines that once AI is described as having an IQ above 140 or 150 and outperforming PhDs in almost every specialized field, what standard will humans use to define the next level of “better” or “stronger”? Extreme pessimists therefore say that “humans are merely AI’s tokens,” nourishment for the model’s growth.
赵丽娜 rejects the premise that humans must compete head-to-head: “Why compare ourselves with a robot? We should compare humans with humans, and ourselves with ourselves.” Machines may go far beyond 100 points, but competitive sports still retain the value of watching humans turn “their own 100 points into 101.”
10. Sports Technology Is Moving From Equipment Design Into Biological Science
Athletic equipment already uses AI extensively. 庄明浩 says shoe technology once focused mainly on materials, but now extends into structural design, biomechanics, renewed analysis of sports data, and algorithm-assisted design.
The larger field is AI for Science: models can simulate protein generation, support drug development, or study how to help athletes eliminate lactic acid, expanding possibilities that were impossible within the limited experimental environments of the past. Biological science is one of its biggest destinations.
赵丽娜 consequently asks whether anti-doping centers, which currently work with what already exists in their databases, might one day use AI to generate more possibilities beyond those databases. This is not a factual conclusion offered by the show, but her concern about technological development and anti-doping screening; 庄明浩 also notes that such databases may eventually need to use AI.
They believe the boundary must at least be attempted, while nobody pretends it will be easy to draw. Humanity previously discussed these rules only in science fiction; now that the technology is arriving, the Olympic system must decide “left or right,” with every choice affecting fairness.
11. F1 Has Already Acted Out the Human Paradox of a Technology-Open Arena
赵丽娜 imagines that the future may bring 2 parallel competitions: one form of “pure sport” restricting both doping and AI technology, and another fully opening the field to robots and every kind of technology so that participants can directly compare technological limits.
庄明浩 believes F1 is already close to the second model: F1 places teams, races, data, systems, and weather inside a technological framework, with weather forecasts accurate down to the minute to determine tire strategy in showers, thunderstorms, or heavy rain.
The question then falls on the driver: when he is operating the fastest land-based engine in the world, what human value remains? 君娴 cites the character played by Brad Pitt in the film resisting data analysis and strategy, arguing that this agency is a performance of “classical aesthetics” but precisely what allows audiences to connect with it.
12. Commentary Is Easy to Automate; Voice and Officiating Remain Contested
Commentary has a clearly automatable chain: image recognition, match statistics, and real-time feedback can be handled by separate modules, and 庄明浩 believes none is particularly difficult in isolation. 赵丽娜 also points out that the 贺伟-style use of classical poetry and elevated interpretation may be something AI handles well.
詹俊’s voice has already been widely used in FIFA games. 赵丽娜 thinks voice may be a commentator’s asset, but 庄明浩 notes that products such as navigation systems have long licensed various celebrity voices, meaning voices have already been commercialized.
Famous commentators can continue to monetize their labels, timbre, and personality signatures, but younger commentators face a higher baseline. As with Premier League players, basic ability must reach at least 80 points before there is any chance of clearing the threshold jointly raised by machines and platforms.
Rehabilitation therapists may soon be replaced by AI, while other parts of the chain will also face varying degrees of penetration. Refereeing is especially conflicted: Hawk-Eye should theoretically improve accuracy, but errors and controversy are also considered important artistic elements of sport; the sporting world has no consensus that 100% rationality is necessarily better.
13. Athletes’ Data Is Turning From Training Material Into a Productive Asset
赵丽娜 distinguishes publicly available height and weight from more sensitive biological data: if systems can continuously read physiological information in the future, who protects it? She raises a further hypothetical concern—could biotechnology use such information to reproduce “another Kobe, another Jordan”?
庄明浩 uses sports video games to illustrate the expanding granularity of data: early versions of Winning Eleven might have given players only 6 or 7 attributes; today, a player in FIFA may have roughly 60, updated each year based on form, tactics, and real-world performance.
When a player controls one athlete, how teammates and opponents move, why the ball flies at a particular speed, and whether the goalkeeper can save it are all algorithmic questions. Sports games have used AI for years; they simply were not packaged in the current large-model narrative.
赵丽娜 would rather exchange data for personal improvement: AI provides the correct answer first, and athletes train with both the question and the answer instead of relying solely on thousands of repetitions to build muscle memory; everyone can also receive a customized coach. The everyday version is photographing food and having calories calculated automatically, with convenience and privacy authorization arriving together.
14. Compute May Concentrate, While Open Source Spreads Capability Back Outward
赵丽娜 worries that AI competition may shrink into a game between China and the US, or even between a small handful of elite people in those 2 countries, with “intelligence, capital, and resources” sealed inside a narrow circle. Those affected would include not only children in China’s mountain regions, but poor regions and countries more broadly.
庄明浩 acknowledges that training costs are astonishingly high, but believes extreme concentration will summon a counterforce. The name he gives it is open source: unlike ChatGPT-style closed technology, open-source systems disclose the model, training methods, internal architecture, data structures, and the way GPUs and data centers are assembled.
DeepSeek’s impact, in his telling, comes not only from opening the model but from opening the training methods as well. People with a certain baseline of capability can copy and use it without an exceptionally high barrier, creating a force that pulls back against compute concentration.
On children in mountain regions, 赵丽娜 worries that AI’s compression of soccer could eliminate an employment outlet; 君娴 offers the opposite possibility: free models and mobile internet are narrowing the information gap, while AI’s replacement of cognitive labor may give physically healthy children new opportunities through physical work. The show also discusses taking over local crafts such as ceramics, with the harder question being how to make children genuinely willing and happy to persist.
15. As the Information Gap Narrows, the Gaps in Asking and Self-Knowledge May Widen
赵丽娜 first lists the abilities to ask questions, understand oneself, raise doubts, and find happiness; 庄明浩 adds aesthetic judgment. 君娴 points out that measuring, training, and sustaining these abilities over the long term are all difficult.
Faced with a model that “knows astronomy above and geography below,” many people simply do not know what to ask. Complex tasks require users to explain the setting, roles, boundaries, goals, and references, so prompts grow longer; most users, however, remain stuck at the first step of “asking about the weather.”
庄明浩’s metaphor is the sun: the model is an infinite source of heat that is always present, and differences in capability come from who can use a conduit to direct that energy toward themselves. 君娴 summarizes it this way: “The information gap is getting smaller, but the gap in the ability to ask questions will get bigger.”
AI may also push competition toward “cultivating the mind and the body.” 庄明浩 recalls a company team hike: the next day required walking 20 kilometers and climbing roughly 2,000 meters; at 1.85 meters and around 220 jin, he and another teammate spent 15 hours in the mountains before getting out. Every step had to be watched closely, and that persistence and the memories that belonged to them could not be experienced on their behalf by a model.
16. AI Can Produce Information, But May Not Build Long-Term Relationships
庄明浩 says AI music can now generate work whose melody sounds fine under almost any demanding brief; film songs, advertising scores, and similar commercial tasks are broadly usable. Music is made of notes, scales, and tones, making it in some sense a finite-rule digital game.
Short video breaks stimulation down into 3, 5, 7, or 20 seconds. 赵丽娜 thinks people may seek other content after consuming enough of it, but 庄明浩 warns that this applies only to some; others will remain in it indefinitely. He calls the condition “internet opium.”
Information-driven podcast use is similarly compressible: a 50-page English report can be handed to Doubao and turned into a 7-8-minute audio briefing. If the task is merely organizing, searching, and summarizing, AI “beat humans flat a long time ago.”
But listening to a podcast may also mean listening for companionship, familiarity, atmosphere, and suspense. Pauses, breaths, emotion, and even sobbing in a voice can bring people closer; 赵丽娜’s summary is the sharpest: “AI can take your timbre, but it can’t take your memories.”
17. Tool Gaps Are Limited; Scenario Boundaries and Trust Matter More
庄明浩 believes the overall differences among leading domestic products such as Doubao, Kimi, and Zhipu are smaller than people imagine. DeepSeek may be slightly stronger at pure copywriting and rhetoric, but it is a difference between “85 points and 87 points”; Tongyi is commonly used for podcast transcription, while removing verbal tics, segmenting, adding timestamps, and tagging require specialized tools.
If someone watches an NBA game, turns off the commentary, and asks AI to help interpret the match in real time, current tools are still inadequate; if the request is only for a player’s recent form, club information, and existing public data, most products can handle it. The capability boundary depends on whether the information already exists and whether it can be accessed in time.
庄明浩 cites AI Talk: Kobe vs. O’Neal, in which AI has the late Kobe say he is “teaching angels to play basketball in heaven,” while weaving shared memories such as 4 a.m. workouts and Shaquille O’Neal’s free throws into the exchange. The scene could never happen in reality, but the characters’ labels and emotional connection with the audience make it feel entirely natural, explaining the appeal of digital relatives.
Asked how to solve data contamination, 庄明浩 refuses to manufacture an answer: even the top scientists have not solved it, so ordinary people need not get trapped in abstraction—“go for a run.” AI fortune-telling can be viewed as deriving conditions and probabilities, but responsibility remains with the user: consult it, but do not take it too seriously. “You still have to live your own life.”
18. Top-End Expression Still Requires Handwork; Humanity Ultimately Has to Solve Itself
庄明浩 used no AI tools for his PPT, summarizing his resistance with a joke: “As long as my PPT is long enough, I can beat everyone.” A standard task such as explaining how dinosaurs emerged can be completed in one click for his son, but his own chapters, images, emotional arc, slogans, jokes, and callbacks all had to be done by hand.
He compares the difference to a factory-line car versus a work made by “a worker hammering it out one blow at a time.” An ordinary weekly report does not need to be handmade, but expression at the top of the pyramid depends on a continuous personality embedded in the details. It is the same thing elite athletes insist on preserving, another “last ground to hold.”
赵丽娜 ultimately moves from pessimism toward limited optimism: humans may not be able to control AI, and do not need to compare themselves with it everywhere, but they can decide whether to depend on it and how to coexist with it. Her conclusion is that “the most important task is to solve ourselves”—which is why sports, the body, and active choice may remain among the parts humans can hold onto to the end of coexistence with AI.