Summary

Zhang Jinjian’s core view is that the world is moving faster, becoming harder, and splitting apart more rapidly—not because information is scarce, but because human attention is out of control. The human eye can take in roughly (10^9) bits per second, while the brain can process only about (10^6) bits and actual action output is just 10 bits; when people give up independent perception and follow others’ actions, the world starts to howl like a microphone placed near a speaker: crowded assets get even more crowded, while the unvisited zones grow colder.
Lack of attention is not, by itself, an investment thesis; a genuine left-field opportunity requires both taste on your side and a market that is not paying attention. Responding to 曲凯’s question that “no one is watching” may simply mean “there is no market,” Zhang cited embodied intelligence: from June to September 2023, while peers warned that companies they had backed in 2015 still had not broken even, Oasis invested in nearly 10 companies in the space. His principle: “If no one in the industry is paying attention to this, that is precisely where we should direct ours.”
The long-term form of an Agent is not an intern inside a computer, but a closed loop combining sensors, full-spectrum models, and embodied intelligence. The frequencies humans can perceive account for only roughly one-billionth of the observable spectrum, while data from X-rays, gamma rays, brain alpha waves, and more remain largely unused; “we’ve run out of data” is true only within humanity’s existing text, image, and audio modalities. The real opportunity may lie in touch, temperature, smell, and many more sensors.
In the AI era, organizational relationships will invert from “AI fits into my workflow” to “I fit into AI’s workflow.” Zhang imagines a future work group with a CEO, CFO, chief product officer, and a dozen other Agents—but only one human. What people retain as their core value is not compute, but articulating real needs, supplying taste, and writing high-quality prompts: “AI bro, you take this Research project. See what I can do.”
Human limitations are not purely a disadvantage; they are precisely what generate strategy, taste, and a commitment to facts. The brain predicts the next input to save computation, but when reality diverges from the prediction, many people refuse to recalculate and dress wishful thinking up as idealism. Zhang’s counter-definition is that “sticking to the facts is actually hard work,” because it requires recomputing from the present.
The foundational pieces most worth revaluing over the next five years, in Zhang’s view, are AR/VR, blockchain, and Agent-native transaction and trust networks. Once the number of Agents equals or exceeds the number of humans, buying wine might involve a butler Agent retrieving memories, a finance Agent allocating 30% of next month’s RMB10,000 budget, and the Agents on both sides negotiating automatically. He believes this identity, information, and trust layer will not continue to be built on the traditional internet.
Intelligence will spread like electricity; the most meaningful startup opportunities may not belong to “large-model power plants,” but to the next generation of products that embed intelligence in specific contexts. Figure’s valuation rose from roughly $800M in October 2023 to about $30B in May 2025, showing how violently consensus can accelerate once it forms in an era of divergence. The cost of individual experimentation is rising accordingly: “When no one is paying attention to you, you have to pay attention to yourself.”

Deep dive

1. The World’s Fragmentation Starts with an Attention Crisis

Zhang Jinjian is continuing the “signal and noise” theme he began in 2018, now in its seventh year. Last year’s topics were “the world is moving faster” and “the world is becoming harder”; this year, he sees accelerating fragmentation—one presidential statement can move global equities or directly alter a classmate’s visa status.
He reduces humans to a signal system of input, processing, and output: perception at roughly (10^9) bits per second, equivalent to a 480P, two-minute video; the brain processing about (10^6) bits, roughly 60–80 pages of a novel; and actual action output of only about 10 bits, like attaching a single label to the world.
Attention is therefore not spiritual rhetoric but an active filter. If a founder decides that a partner is “terrible,” information consistent with that judgment will continually be retained while other strengths and even business signals are discarded, leaving the conclusion that “this guy really is terrible.”

2. Focus Raises Resolution; Defocus Creates Market Howl

Zhang uses the Transformer to explain focus: “Attention Is All You Need” was the breakthrough that gave machines Attention. A good founder’s scarce ability is similarly not to traverse all available information, but to know when to focus on the principal contradiction.
He observed that 梁文锋 “speaks very slowly, listens very carefully, and takes a long time to emit a token,” interpreting the slowness as extreme concentration. Zhang himself does not carry a phone when speaking with founders for an hour: the more concentrated the attention, the finer the information grain. “Like a laser, it shoots farther.”
Once the world speeds up, the system starts to overload. Podcasts update hourly; Trump may post a message every hour, contradict himself, or delete it. Conversations in the secondary market have also deteriorated from “What is your logic and what is the trend?” to “Just tell me the wealth code first.”
When inputs simply follow other people’s outputs, it is like placing a microphone near a speaker and creating feedback: what is hot gets hotter, while cold areas are ignored. Zhang’s contrarian view is that “the world has more opportunities, not fewer”—provided people rebuild independent perception.

3. Early Bets on Embodied Intelligence Came from the Unoccupied Zone, Not Consensus

In March or April 2023, Zhang said: “AI is software. There must be hardware to carry it. I don’t know what that hardware is called yet, so let’s call it a humanoid robot for now.” His colleagues instead thought he was saying this because he would not let himself research AI; one even announced on the spot that he was resigning.
After the team’s research, one industry veteran pleaded with them: “We invested in this back in 2015 and still haven’t broken even. Whatever you do, don’t come in.” Oasis reached the opposite conclusion: the timing had arrived. From June through September 2023, it invested in nearly 10 companies now classified as embodied-intelligence companies, which Zhang says later became leaders in their respective fields.
This was not a mechanical strategy of buying whatever was coldest. Attention without taste can simply be a bubble from your own perspective; with neither attention nor taste, it is “annihilation.” The real edge comes from the high-resolution judgment formed through sustained focus.

4. Lack of Market Attention Can Be Protective—but It Needs a Time Limit and Validation

曲凯’s question was direct: Could places nobody is watching simply be places nobody needs? If a founder has strong taste but the market withholds both attention and commercial returns for years, what should they do?
Zhang answered with Nvidia stock: if you are certain the value is real, the longer the stock stays cheap, the more time you have to build a position. Fear of being ignored often means you “do not believe in your own taste enough.” When conviction is real, more attention should make you more anxious; less attention gives you more time.
He cited a researcher-founder from a poor family who completed his undergraduate, master’s, and doctoral degrees at top schools, then chose foundational research that attracted no attention and was not even licensed by the state. Fewer than 10 people persisted for seven years, betting that the market would start in 5–10 years. In 2023, the global market issued four licenses; he secured one and later won roughly half of all orders worldwide.
Persistence is not an excuse to wait forever. Everyone should define their own window, whether three years or 18 months. Oasis’s internal rules include refusing to discuss products with only a six-month shelf life, as well as products planned solely because market attention is high.

5. Rebuilding Attention Starts with Admitting What You Actually Want

Zhang traces the filter back to “knowing yourself and loving yourself.” One founder cared intensely about making money but considered talking about money low-class, so he spent every day preaching morality, ideals, and the future to colleagues and investors. The problem was not wanting money; it was refusing to organize his attention around his real objective.
If the goal is to make every colleague like you, you can make decisions around that goal—but you must accept the side effects, including possibly making no money and accomplishing nothing. “What do I actually want?” is not a value judgment; it is the precondition for the filter to function.
AI is pushing loss of control to an extreme. In the past, if Cristiano Ronaldo did not play, there was little new content; now AI can generate infinite video, images, news, and analysis. Zhang expects people with and without independent attention mechanisms to sort themselves naturally over the next two years: some will lose their views or become extreme, while others will become quieter and more comprehensive.

6. Humans Live in an Extremely Narrow Spectrum; Emotion Aligns Five Modalities

Zhang describes the observable electromagnetic spectrum available to humans as roughly (10^{24}) in scale, while the range humans can perceive is about one-billionth of that. The same 张津剑 as a dog would smell more odors; as a fly, he would see different images. “The world is not limited; the information we can perceive is.”
Civilization expands along one path by using MRI, X-ray machines, weather maps, and gravitational-wave detectors to widen perception; another path is cross-modal recording, such as turning sound into images with musical notation or storing sound on tape. Humanity accumulates the “infinite” by continually expanding into and preserving frequencies it could not previously access.
Once the five senses enter the brain, what truly synchronizes them is not a single logic but feeling or emotion. A potential partner’s résumé and views may all check out while your body keeps feeling uneasy. Zhang believes that multimodal result should not be overridden by consciousness: “100% failure” is his deliberately strong judgment, but the point is not to ignore the feeling that something is wrong.
What people ultimately preserve is often a feeling. When an entrepreneur becomes wealthy and searches for a first love, he may not be looking for that person so much as the feeling of being in love for the first time. Meeting the person but failing to recover the old feeling shows that the multimodal information in memory has long since been recoded by emotion.

7. The Agent’s Frontier Is Full-Spectrum Perception, Not Just Language

Once language tokens become the center of the modality stack, Zhang sees today’s DeepSeek language models and MiniMax audio models as only the beginning. Future systems will include touch, temperature, and smell models, processing information across gamma rays, X-rays, ultraviolet, infrared, and other frequency bands.
“There is not enough training data left” is therefore, in his view, a premature conclusion. Humanity has mainly consumed the limited data in text, sound, and images; X-rays, gamma rays, brain alpha waves, and many other datasets remain largely unused, while vast amounts of real-time information have not even been given a use case.
A complete AI system should combine three parts: sensors, full-spectrum processing by Agents, and embodied action. The world it perceives may be as unimaginable as the human world is to an ant. Token costs, whether Chinese founders can build it, and whether the end state is called ASI, Super AI, or AGI are not the questions Zhang considers most important.

8. Workflows Will Center on AI; Humans Will Supply Needs and Taste

Zhang rejects treating AI merely as a Deep Research, Copilot, Sourcing, or Research tool: “We are not folding AI into our workflow; we are folding ourselves into AI’s workflow.” A future group chat might contain a CEO, CFO, chief product officer, and a dozen other Agents, with perhaps only one human.
Power consumption reinforces the inversion. The human brain uses about 20W; an A100 uses about 400W. Biology’s value to AI may not only be that AI helps humans develop drugs, but that biology provides low-power capabilities to AI. The colleague who says “please” every time he speaks to AI is also expressing a change in mindset—and respect for AI.
The human’s first job is to articulate a real need; the second is subjective forecasting. The brain predicts the next input and skips computation when the prediction matches. When reality diverges, it should recalculate, but many people refuse and beautify wishful thinking as idealism. “Sticking to the facts is actually hard work,” because it requires continuously updating the model.
Limited time, resources, and attention force humans to develop strategy; forecasts that remain effective over time are called taste. AI’s infinite nature, by contrast, leaves it lacking in taste. Zhang therefore believes prompts will become “very, very valuable,” while taste comes from passion: lipstick, art, structure, or protocols can all become distinctive human contributions.

9. Accepting Your Limits Works Better Than Forcing Yourself into the “Right” Person

A founder with a consulting background failed after two years and was then told by a doctor that he might have only two weeks to live. His review concluded that he had “spent his whole life talking and not doing.” After recovering, he decided to continue “talking and not doing,” and spent a year finding a CEO who “only did and did not talk.” The company went public four years later; even outsiders did not know he was the actual controller.
The founder distinguished between two kinds of review. A future-oriented review always wants to change; a death-oriented review can only accept. Once he fully accepted himself, he stopped compensating for his personality and built a team around his strengths.
When he cannot see himself clearly, Zhang recommends treating his own problem as someone else’s, then simulating how higher-level decision-makers such as 梁文锋 or 闫俊杰 would answer. He then asks them directly. If the prediction matches, it is a successful distillation; if not, he asks why and gets one round of SFT. “Life is an RL plus SFT process.”
There is no universal standard for what deserves focus; the only question is whether it serves the current objective. Zhang cites the numbers that purposeful work increases brain energy consumption by only about 5%, while purposeless global scanning increases it by roughly 15%–20%—three to four times as much.

10. The Agent Economy Five Years Out Needs New Interfaces, Trust, and Leadership

Zhang believes there is no need to wait 10 years; significant change should arrive in roughly five. AR/VR will be one way humans interact with new information, while blockchain is “severely underestimated.” The number of Agents may eventually equal or exceed the number of humans, and they will not live on the internet so much as in a distributed world built on blockchain.
The red-wine example sketches the chain. A butler Agent retrieves personal history and preferences; a finance Agent sees that next month’s budget is only RMB10,000 and allocates 30% of it; another Agent searches the entire internet. The seller is also represented by an Agent, while blockchain transmits information, identity, and trust between the two sides. Data an Agent cannot collect, or work it cannot complete, is handed to embodied intelligence.
Humans will eventually split into two groups. People with taste will direct dozens of Agents and make their taste available to more people; those whose attention has collapsed will be taken over by the information stream. Today, you still have to open Douyin; in the future, one turn of the eye may cause VR to fill your field of vision with the relevant content. “The future information stream will be passive.”
Relying on high-quality signal sources is not itself a mistake. The key is whether you are making an active decision or reacting passively. Zhang can judge that 曲凯 is more professional and therefore entrust an event to 42章经. But if he simply watches every move others make and follows, his own perception system is still not working.

11. Intelligence Will Spread Like Electricity; Opportunity Belongs to Context and Inner Drive

Intelligence powers the brain; Zhang believes the energy that powers attention is “heart force.” The Industrial Revolution moved people from physical labor to intellectual labor; the AI era will move them again toward work powered by inner drive. A founder’s differences ultimately show up in persistence during hardship, restraint during pride, courage, resilience, and self-evolution.
He recommends not looking at your phone during the first hour after waking, or at least the first 10 minutes. Heart force is recovering and attention is rebuilding; exposing yourself to the information stream too early is when you are most likely to be taken over by the outside world.
In Zhang’s analogy, large models and cloud companies are power plants and grids. Intelligence will eventually be priced by token, perhaps even with tiers and time-of-use pricing like electricity. But the meaningful startups may not be “State Grid”; they may be the radios, washing machines, and refrigerators of the past—companies that align intelligence with specific contexts. The future will produce companies analogous to Gree, Midea, and Xiaomi.
Fragmentation raises the cost of failure while shortening the path to success. DeepSeek was still little known before the 2025 Lunar New Year, then quickly became a household name. In October 2023, Zhang said Oasis was among the earliest and, at the time, one of the few investors in China paying attention to embodied intelligence. Figure’s valuation rose from roughly $800M in October 2023 to about $30B in May 2025; Oasis did not invest because it was focused on China. Zhang’s conclusion: “When no one thinks highly of you or takes you seriously, you have to take yourself seriously.”

[AIEWF Preview] Gemini in 2025 and Realtime Voice AI AI in China ⑤: Making To B Deliver—RMB30M Without Moutai