Summary

Over more than a decade, 测测 shifted four times—from tools to a two-sided platform, then AI, and now embodied companionship; its existing business now provides the cash-flow base for robot experimentation. It started in 2011 with tools for astrology, MBTI and relationship analysis, launched a paid human-services platform in 2016, began planning for AI in 2017, released its first conversational robot in 2019, and started building hardware last year. 任永亮 says the platform has 50 million registered users, roughly half of whom have used AI, with growth mostly ranging from “a few dozen percentage points to 100%” over the long term—but it has never experienced explosive growth.
General-purpose models have lowered the barrier to single-point features, but they have not eliminated the room for vertical products in context, professional supply and delivered outcomes. After going through three consecutive “New Year’s anxiety” episodes—ChatGPT, Sora and DeepSeek—任永亮 concluded that 测测 needs to combine long-term user information, professional tools, human services and psychology-labeled data, fine-tune on a foundation model, and bring in external information to move users closer to a delivered result. “Every year brings a feeling of panic,” he says, but the underlying business continues to grow.
The robot is a hardware bet funded by a mature business, kept within a safe investment range, yet difficult to explain through conventional business rationality. 任永亮 admits, “My decision to innovate was not rational; it was purely emotional.” Both his team and hardware veterans warned him about the risks, but he still rejected a relatively lucrative acquisition offer, partly to preserve cash flow and control over decisions. The project has not yet raised external funding; failure may mean only that “the financial statements won’t look quite as good.”
The first-generation product targets active companionship at home—not housework or K-12 score improvement—with interaction quality and emotional value as the core variables. It is designed to initiate interactions based on context, but “proactive dialogue must not become proactive harassment.” Sensitive video data will stay on-device, while sensors, motors and local compute mean the hardware will not be cheap. 任永亮 sees roughly 10,000 units as an early experimental scale, far below the “10 million units to create a category” threshold cited by a friend.
测测 serves highly non-standardized emotional needs that are difficult to scale but persist over time, and its user base has already formed a clear profile. Women account for about 80% of users, as do users in China’s first- and second-tier cities; relationships are the biggest source of confusion. AI has more users than human counseling because it is cheaper, easier to use and viewed by some users as more objective, but human presence, eye contact and hugs become scarcer—not less valuable—in the AI era. “Happy families are all alike; unhappy families are each unhappy in their own way,” which is precisely why the industry remains fragmented.
任永亮 frames companionship products through a “dopamine–oxytocin–endorphin” model: instant feedback attracts use, a sense of relationship creates connection, and overcoming difficulty produces growth. He does not want to replicate short video’s “maximize dopamine” addiction loop, but wants robots to interrupt the loop and become part of daily life. Big eyes, hugs and tactile design are meant to trigger a caregiving instinct, while education should help users cross a threshold of discomfort. The commercial challenge is therefore to make the product attractive enough without turning companionship into another endlessly stimulating machine.
His macro view of AI is more pessimistic than his view of 测测: AI may destroy the income and dignity created by the division of labor without creating an equivalent number of jobs. He therefore defines robots and broad psychological services as “spiritual shelters,” while counseling, companionship and empathy may become sources of value increasingly dependent on “the human being himself.” This is the deeper logic behind the hardware bet: if tool-like AI cannot form a closed loop, embodied relationships may be a “narrow gate” for amplifying AI’s value.
What runs through his entrepreneurial history is not a deep belief in astrology, but a science-trained mind’s long-running fixation on people, cycles and computability. From using support vector machines to predict siRNA sequences in 2004, to applying chaos theory to understand initial conditions at birth, to asking whether information can be added to the mass–energy equation, 任永亮 has kept searching for the point where technology meets human beings. He calls the robot “an interesting experiment”: even if it fails commercially, he wants to use the company as a vehicle for innovation and keep asking whether “this world is ultimately computable.”

Deep dive

1. 测测 Was Born from Emotional Confusion, but Its First Job Was Explaining and Consoling—not Predicting Fate

任永亮 can no longer tell whether the story that he started the company “because a girl said their star signs were incompatible” is the original truth or a version hardened through repetition. What he can confirm is that, after running into relationship trouble, he “went to the doctor in a panic,” searching for a framework to explain his confusion and ease his fixation.
His medical background gave him another perspective: before penicillin, much of medicine was also about comforting people. He cites the epitaph of a tuberculosis doctor—“To cure sometimes, to relieve often, to comfort always”—and believes emotional consolation has never been merely an add-on to medicine.
He also mentioned 测测’s collaboration with the Center for Narrative Medicine at Peking University Health Science Center: how patients tell the story of an illness changes how they understand its meaning and affects the doctor–patient relationship.

2. From Preventive Medicine to Machine Learning, He Chose Computation over Clinical Practice Early On

任永亮 studied preventive medicine at Peking University Health Science Center, focusing on epidemic control, nutrition and sanitation, maternal and child health, and hospital administration; surgery was not part of the curriculum. A year after enrolling, he encountered computers and began writing code almost every day, while often cramming for his professional courses just before exams.
In 2004, he remotely joined a US professor’s bioinformatics research, using the machine-learning method of the time—support vector machines—to predict siRNA sequences, and may have published a paper as first author. He originally wanted to make a mark in science, but later decided that “adding a tiny piece to this building was enough,” which made it easier for him to leave and start a company.
His medical and research experience took him to IBM, where he worked on foundational hospital-informatization projects. But IBM’s To B orientation made him realize that he wanted to affect a large number of users directly. Looking back, he notes that IBM’s healthcare AI is often cited as a case of “heavy investment, relatively little output.”

3. The First BP Already Envisioned Personalized Recommendations, but the Entry Point Had to Be Frequent and Small Enough

The first 测测 BP defined the product as a “big-data-based personalized recommendation engine”: combining personality, birth information and demographic data to recommend personalized daily-life advice. It was not a news recommendation product, but a system for advising people on their own circumstances.
Astrology, psychology and tests were not chosen simply out of personal interest. Specialized medical content had a small audience and high education costs, while a To C founder without resources or outside backing needed something users were willing to discuss frequently to support services they used only occasionally.
任永亮 also tested demand through social-network plugins and various small products. His summary of why this market suited him was simple: “Top programmers generally look down on it,” while people who liked this content were not necessarily able to write programs. That created an early opening for him.

4. When Technology Could Not Make Holistic Judgments, 测测 First Became a “Buffet”

When he started the company in 2011, 任永亮 already wanted to build AI Q&A: combine the many conclusions generated by psychology and astrology systems and answer users’ specific questions directly. Traditional machine learning could not perform that integration, so he retreated to a tool-based product.
He compared the early product to a “buffet”: daily scores, astrology, MBTI, relationship analysis and other viewpoints were presented as objectively as possible, leaving users to select what was useful. 测测 consequently became known as relatively comprehensive. It later found that most of its core tools were closely connected to Jungian psychology, leading the team to treat Jung as its “patriarch.”

5. Four Stages of Evolution Gradually Changed How Personalized Services Were Supplied

The first stage combined tools and community. If a piece of psychological or personality content could be digitized and productized, the team tried to build it. Names such as “Birth Code” and “Wisdom Cards” deliberately softened labels associated with practices such as Bazi and Tarot, allowing users to approach them first as entertaining tools.
The second stage began in 2016. Professionals and ordinary users coexisted on the platform, and as audio and online payments matured, 测测 borrowed from the paid-knowledge model to build a system for talent testing, counseling and paid matching.
Human services could only be delivered one-on-one, placing hard limits on efficiency and supply. The third stage therefore shifted to AI: the team began planning in 2017 and launched its first conversational robot in 2019, with the immediate goal of solving the shortage on the supply side.
The fourth stage began last year: giving broad psychological services an embodied form, so the product would no longer wait for users to ask questions but could enter the home, respond to specific situations and initiate interactions. 任永亮 believes pure software remains a passive tool, while an embodied product might become a companion in daily life.

6. AI Usage Has Surpassed Human Counseling, but the Two Forms of Supply Are Not Simple Substitutes

AI has more users primarily because human counseling is expensive while AI has a lower barrier to entry. Some users also worry that people are unreliable and instead view machines as “more impartial and objective”—a view 任永亮 specifically limits to only some users.
The longer he works on AI, however, the more valuable he finds human supply. A person’s gaze, a hug, or even “two people simply sitting together” comes from embodiment and feels different from a virtual service. Platform-wide and practitioner-related metrics are still growing as well.
This creates a two-track bet for 测测: use AI to expand low-cost personalized supply while retaining the relational value created by human services. The embodied robot is an attempt to give machines some measure of real presence.

7. General-Purpose Models Bring Repeated Anxiety; Vertical Opportunity Comes from Longer Context

任永亮 says he has “a feeling of panic every year”: ChatGPT in 2023, Sora in 2024 and DeepSeek in 2025. The rapid pace of progress made several Spring Festivals feel “deeply unsettling.”
His calmer conclusion is that general-purpose models answer immediate questions, while vertical products construct a longer context for users, combining professional tools, historical information and different services through something like MCP to deliver a more concrete result.
DeepSeek may put more direct pressure on the foundation-model industry, while other competitors face greater pressure; products such as Yuanbao can also work with it and use it to expand their user bases. 任永亮 therefore does not claim the threat has disappeared, only that this is a “gradual cooling process.” The startup world is also shifting its focus from model capability to what value agents can actually deliver.

8. The Real Vertical Moat Is a Combination of Data, Tools and Services—not a Single AI Astrology Feature

Faced with the flood of AI MBTI and AI astrology startups, 任永亮 calls them “low-hanging fruit.” 测测 can build these single-point features, but they are no longer where his interest lies.
His competitive strategy is deliberately restrained: the leader does not necessarily need to bear every original risk. “If you want to know how to innovate as the second mover, apply the innovation.” That logic works for online broad psychological services, but does not explain why he later entered hardware voluntarily.
Technically, 测测 fine-tunes a general-purpose foundation model and builds context from years of tools, counseling interactions and user feedback. The team has also hired psychology graduates to participate in professional data labeling. When the host asked whether that was still insufficient, 任永亮’s answer was: “Enough.”
The next step is closer to Deep Research: automatically call information outside the model and break down multiple paths, conditions and outcomes for questions such as “how to make yourself happy.” 任永亮 wants to deliver more than a single conversation—a result tailored to the user’s circumstances.

9. Fifty Million Users Prove the Demand Exists, but Have Not Turned Broad Psychological Services into a Standardized Industry

任永亮 disclosed that 测测 has roughly 50 million registered users, half of whom have used AI; he could not recall the precise number of active AI users. Platform growth has long been high by ordinary standards, ranging from “a few dozen percentage points to 100%,” but he repeatedly emphasized that it was not a sudden breakout.
Women account for about 80% of users, as do users from China’s first- and second-tier cities. The most common questions concern relationships. “A gathering place for hopeless romantics” was not the original positioning, he says, but admits that the description “makes some sense.”
Asked why a widely shared need had not produced a much larger company, 任永亮 borrowed the line that “happy families are all alike; unhappy families are each unhappy in their own way.” Everyone’s experiences and problems differ, human counselors can work only one-on-one, and demand therefore resists standardization and remains fragmented.

10. The Embodied Project Is an “Irrational Decision” Protected by Cash Flow

任永亮 says plainly: “My decision to innovate was not rational; it was purely emotional.” Online broad psychological services had already built a meaningful advantage, and the rational choice would have been to follow competitors. A robot, by contrast, requires software, hardware and services to work together, with an entirely different risk profile.
Company management was not uniformly opposed, but many worried that the team lacked hardware DNA. Friends repeatedly introduced him to hardware veterans specifically to warn him that “hardware is very difficult.” That resistance instead fueled his contrarian impulse and desire to try.
The host raised last year’s acquisition offer from a buyer at a relatively high price. 任永亮 confirmed that he declined it, with the robot as an important reason: he wanted to retain the operating cash flow needed to fund innovation and did not want to give up control over the company’s direction.
The project is currently funded with company earnings and has not raised external capital. 任永亮 believes the investment remains within a safe range; failure might mean only that the financial statements look less attractive. The product has not entered mass production, and the choice and opportunity cost may become clearer after more time.

11. The First-Generation Robot Will Be a Family Companion; Education Is Simply the First Use Case

The most direct product inspiration came from home. 任永亮 often saw his wife lose her temper while helping their child with homework and wanted the robot to create value in education first. But it is not aimed at improving K-12 scores; it combines education and psychology to serve the emotional needs of the family.
He believes the most important innovation in large models is interaction, while household manipulation is further from maturity. A company that does not rely on concept financing should first build something that can actually land. The goal is therefore “the first robot in the family,” focused on autonomy and proactive interaction rather than an all-purpose body.
The robot will initiate dialogue based on context, but “proactive dialogue must not become proactive harassment.” It will power on automatically when taken out of the box, while users can choose whether to turn it off. The key is not how often it speaks, but whether it knows when to come closer and when to stay quiet.

12. On-Device Privacy, Mobility and Price Form a Triangle with No Easy Compromise

The most sensitive data will stay on the device, and video will not be uploaded to the cloud. That requires local chips and compute. Add sensors and motors, and the robot must both understand its environment and move through it; the cost therefore cannot be low by default.
任永亮 mentioned a Japanese companion robot priced at around RMB30,000, which the team says is essentially sold at cost. 测测 has not disclosed its price; he only said that it would also be “very expensive,” but not so expensive that deployment becomes impossible.
The plan is to launch the product “next year.” 任永亮 describes hardware development as a continual process of agonizing trade-offs: intelligence cannot be too limited, movement cannot be removed, and costs must still be contained. More than half of his own attention is now focused on the project.
One friend believes hardware sales must reach 10 million units to create a new category. 任永亮 says he certainly cannot achieve that. Following the logic of Crossing the Chasm, the first phase will target early users, with an internal expectation of around 10,000 units—enough to complete the experiment.

13. The Design Language of Companionship Breaks Down into Three Different Reward Mechanisms

任永亮 roughly decomposes love into dopamine and oxytocin: the former corresponds to passion and instant feedback, the latter to affection and durable connection. He emphasizes that this is only a material-level framework developed for robot design, not a complete definition of love.
The central question he took from the LOVOT team was how to use big eyes, movement and hugs to stimulate oxytocin to the right level. Pets and stuffed toys were also early reference points: soft touch and the feeling of being cared for allow a relationship to exist beyond the screen.
Endorphins correspond to delayed rewards such as running and studying: users must first overcome discomfort and cross a threshold before pleasure arrives. A game’s dopamine loop is short, while homework requires difficulty first and may offer no immediate feedback.
He does not want the robot to copy short video’s “maximize dopamine” product logic. The ideal relationship is to make someone who is heartbroken or anxious “less miserable” first, then build a long-term connection through support, understanding and dependence, while preserving the interruptions and difficulties necessary for growth.

14. MBTI Is Not a Fixed Label for Him, but a Low-Cost Coordinate System for Understanding People

任永亮 says MBTI “is not a mirror of you”; it is closer to a definition of your current state and includes an element of active self-understanding. It differs from the more scientifically oriented Big Five personality model, and there is nothing wrong with a type changing.
He uses a computer analogy for the four dimensions: I/E describes whether you recharge through solitude or interaction; N/S whether you focus on the abstract or the concrete; T/F whether you process information through reason or feeling; and J/P whether you bring judgments into the world through order or freedom.
INTP gives him a positive signal that he is suited to innovation and entrepreneurship. A company executive once said he could not stand 任永亮’s constant changes and flights of imagination, while admitting that the traits were well suited to starting a company. He relayed outside analyses classifying Jack Ma and 张一鸣 as INTP, though neither classification came from a self-administered test; Lei Jun, by contrast, has been analyzed as a rare E-type entrepreneur.
He also mentioned a party-school research report that found ESFJ relatively common among civil servants, almost the opposite of INTP. That made him more comfortable with his unsuitability for order-driven careers. The team is also working on “测测 MBTI,” inferring the likely type of children who are unable to answer the questions by observing them.

15. Astrology and Birth Data Are Cycle Hypotheses in His System, Not Reproducible Science

On birth data, he says the system fundamentally belongs to the solar-calendar tradition and concerns the relationship between the Sun and Earth. He therefore understands these questions as matters of cycles and emphasizes the Sun’s importance to Earth—for example, how changes in distance could alter the planet’s environment.
He uses chaos theory to explain the limitation: complex systems are highly sensitive to initial conditions, and a person’s birth time, location, genes and family environment are all initial conditions. As with three-body motion and weather, a small difference at the start can produce a completely different long-term outcome.
Astrology, Bazi and similar systems are “macro hypotheses” that can explain only part of reality at most. He cites Kondratiev waves and the 60-year Jiazi cycle, arguing that a generation’s roughly 60-year cycle of wealth creation and depletion can leave observable patterns, but cannot precisely predict an individual’s fate.
Science describes events that can be reproduced under resettable conditions; “life is difficult to reproduce.” Opportunities and economic cycles are also irreversible. These systems are therefore closer to cultural phenomena: they use a small number of cyclical clues to provide a feeling of certainty in an uncertain life.

16. Humans Repeatedly “Test Themselves” Because the Mind Naturally Compresses the World into Concepts

任永亮 believes Homo sapiens do not store information in the raw form of the objective world. We first convert it into theories and concepts to lower the cost of understanding. A monkey performing a visual task might remember many pixels, while a person abstracts a shape into “a circle” or “a person.”
When a system cannot generate purely scientific laws, people still use hypotheses and cultural definitions to describe it. The durability of astrology and personality tests does not depend only on whether they are true; they also serve practical psychological functions by helping people organize experience, understand others and find comfort in situations beyond their control.
The host extended the discussion to large models: when people demand that a model “understand the laws of physics,” they are projecting Homo sapiens’ own requirements for regularity onto machines. 任永亮 believes language reasoning has made breakthroughs, but AI has not truly mastered the physical world. Only discovering new scientific laws would bring it close to a higher peak.

17. Software AI Can Expand Supply, but Embodiment Sets the Ceiling for Psychological Services

In theory, AI can deliver personalized experiences to thousands of users at low cost and ease the labor bottleneck in broad psychological services. But 测测 still needs to keep reconstructing context, scenarios and professional knowledge rather than waiting for general-purpose models to complete the business on their own.
The host’s challenge was direct: if software could itself become a large market, why not concentrate limited resources on doing one thing well? 任永亮 answered that the team has worked on software AI continuously since 2017 and 2018, while a window is now opening in the embodied field.
He believes virtual services and tangible entities provide different forms of value. One reason humans remain irreplaceable is that they retain embodiment—eye contact, hugs and shared presence. If psychological problems ultimately require solutions from multiple angles, the robot is one path he wants to test.

18. Pessimism about AI’s Social Impact Gave Rise to a “Spiritual Shelter”

任永亮 believes the Industrial Revolution relied on the division of labor to professionalize and instrumentalize people, but it also brought most people into the system of value distribution. AI is replacing precisely these tool-like roles; value may accrue to AI, leaving ordinary people unable to benefit from the system in the same way.
He uses AI programming as an example: higher efficiency will not create more programmers and may instead reduce the number of jobs. The work added by training AI is “negligible” compared with the work eliminated. He does not currently agree with the optimistic view that AI will inevitably create more new professions.
He mentioned Altman’s plan and report for giving every person distributable wealth: participants’ health might improve slightly, but they might not become happier, because income cannot automatically restore a sense of competence and dignity. The metaverse is one imagined destination; psychological counseling and companionship between people may be another.
His long-term view is that agricultural society depended on physical strength, industrial society on intelligence, and the AI era may depend more on “the human being himself,” especially emotion and empathy. He hopes 测测 and its robots can become a “spiritual shelter,” helping people adapt to life after the decline of tool-based value.

19. The Robot Is a “Narrow Gate” at the Intersection of Technology, Market and Personal Destiny

Facing competitors with stronger research, hardware and supply-chain backgrounds, 任永亮 admits he can only search for a gap between market and technology. The development of large models and autonomous driving, China’s supply-chain advantage, and his own shift from emotional confusion to parenting needs may together open a “narrow gate.”
测测 has given him software, AI, a services platform and large-scale human interaction—not manufacturing capability. He wants to complete the hardware side and become a “triangle warrior,” while treating proactive interaction as the robot’s most important capability.
He keeps asking, “Why me?” Every step of his past has connected new technology to a specific market, from mobile internet and online communications to AI. He is not sure whether this is an aptitude or a hobby; he calls it his “destiny.”
When ChatGPT exploded, he “deeply regretted” not paying attention to large models during the GPT-2 era, as if he had missed another possible life as a scientist. But he believes a company has become an important vehicle for technological innovation, often with more data and compute than an academic institution. The robot is his experiment for continuing to explore the road not taken.

20. The Final Question Returns from Business to Whether the World Is Computable

The question 任永亮 leaves behind is whether “information” can be added to the framework in which energy and matter transform into each other. To him, this is equivalent to asking whether the world is virtual and whether there is an underlying algorithm that can be computed and expressed.
He has no preferred answer, because measurement must come before computation. To measure the entire universe, one would need to reach the speed of light, and even converting all matter into energy might not provide enough to complete the measurement and prediction.
This brings the entrepreneurial story full circle. From reading Einstein: His Life and Universe in high school and wanting to explain consciousness mathematically, through support vector machines, a broad psychological platform and companion robots, he has kept asking whether people can be understood and simulated. The robot is commercially “an interesting experiment”; the personal value lies in the experiment itself.

Conversation with Xiao Feng: Cool-Headed Thoughts on Hong Kong Stablecoins Episode 148 - July 18, 2025