Liu Yuan on AI Hackathons, Startup Moats and Unease About Change

Zhang Xiaojun speaks with ZhenFund partner Liu Yuan about AI hackathons, startup investing and the emotions accompanying technological change. Originally released in 2023, the discussion reflects that period.

Liu describes years of looking at early-stage projects across industries. Outcomes vary, he says, while curiosity keeps him engaged with healthcare, robotics, education and other fields.

They discuss his reputation as a literary-minded investor. Liu says the relationship between artistic interests and a founder’s practical work is complicated and resists simple labels.

He thinks natural-language interfaces let more nonprogrammers experiment with AI. Expression, vocabulary and judgment about outputs can be useful, without making technical knowledge irrelevant.

The discussion turns to a revival of hackathons. Participants form teams, conceive ideas and demonstrate prototypes quickly, providing practice and exposing observers to many product experiments.

Liu recalls an AI hackathon organized by Jike: rapid presentations were exciting but left investors little time to understand each team. A prototype can demonstrate a possibility without being a durable business.

He observes that many application experiences depend heavily on underlying model capabilities. This raises an investor’s question: how can applications accumulate value if competitors can access similar models?

The concern is not that products are useless, but where value will ultimately accrue. Customer relationships, workflows, distribution and iteration may matter, yet none automatically creates a moat.

Zhang asks how a team’s advantage can endure. Liu says that when many similar prototypes appear, investors must distinguish a short-term demonstration from lasting product capability.

They discuss founders and product transactions in the Jike community. Liu describes a browser extension later acquired by another founder as an example of a small product attracting users and creating opportunities.

He describes a path from student groups, small tools and hackathons toward learning team-building and product development. Both positive and negative feedback can help founders improve over time.

Liu contrasts founding with working inside a large company. A friend’s experience illustrates different responsibilities and rhythms, not a claim that one path suits everyone.

The discussion turns to founder salaries. Liu says compensation should be considered alongside cash reserves, living needs, team structure and company stage; excessive spending can shorten runway.

He acknowledges there is no single standard: cofounders, employees and people with different family obligations face different circumstances.

Looking back at the event, he says investors can tire after seeing many projects in succession. With high volume and similarity, understanding a team requires deliberately making time.

The hackathon showed him that small teams could quickly build appealing demos on existing models. Precisely because prototypes were easier, questions of durable advantage and value accumulation became harder.

Comparing Chinese and US ecosystems, he observes differences in urgency around generative AI, capital and startup conditions. He presents this as a personal observation, not a conclusion about every company or researcher.

He discusses different founder and talent backgrounds, including young technical talent, experienced operators and scientists. Each can bring strengths, while opportunity and environment also matter.

Liu says pursuing an interest requires resources and room to experiment. Family circumstances, education and market conditions affect who can explore for long periods, highlighting unequal opportunity.

Asked whether entrepreneurship can be trained like an exam, the speakers say experience and feedback help, but startup problems change and cannot be solved solely by applying fixed answers.

They also discuss professors and scientists starting companies. Liu says that beyond technical ability, founders need to learn team building, equity arrangements, leadership and market understanding.

He compares expectations across funds and investors, suggesting that some place more emphasis on commercial execution and product skills. Standards vary by institution, stage and project.

For companies built around an earlier generation of AI, he suggests accumulating expertise and value in specific industries rather than assuming an old technical advantage will last forever.

Liu says ChatGPT prompted him to reconsider the boundaries of human work and left him uneasy. He is describing personal feelings and questions, not an established fact that human abilities have been replaced.

He suggests many services involve acquiring, organizing and communicating information and may be affected by AI. Healthcare and education also involve relationships, practice, responsibility and safety and cannot be reduced simply to inputs and outputs.

Asked about the fund’s AI investments, he says they span multiple layers and mentions an interest in experienced repeat founders. A team’s direction can change, so investors must keep observing.

He asks which information tasks remain hard to automate in the near term and acknowledges major open questions about brains, model mechanisms and consciousness. Generated content alone does not establish human-like imagination or consciousness.

Liu distinguishes analyzing information in a business plan from building trust face to face. He also believes investors should investigate how to use AI in their own work.

He describes a founder’s way of identifying opportunities: look for things foundation models still do poorly and things large companies may find difficult because of their own strategic positions. One example is coordinating models from different providers.

A short-lived product can bring users and feedback, but a team must consider whether it can retain value in the next phase. Mistaking temporary growth for permanent advantage carries risk.

He reflects on seeing many projects but investing in few: his bar may be high, or his sample may be incomplete. Scarcity of strong teams does not mean investors’ screening is error-free.

They discuss two sides of the market: broad excitement about AI alongside caution about valuations, teams and sustainability. Agreement on a direction does not automatically answer a specific investment question.

Liu sees uncertainty as an opportunity because change creates new information and needs. Opportunity does not guarantee success; founders still need to test and revise under uncertainty.

He observes that abundant news and events do not necessarily mean an equal number of investable companies. Research projects, individual open-source work and commercial products have different goals and kinds of value.

The investment team is reconsidering whether its bar is too high: it wants to use past experience to avoid mistakes while leaving room to experiment with different teams in a new era.

In the closing questions, Liu discusses literature, film and whether AI can create subtle work. He uses his reading and viewing experiences to express curiosity and unease, without treating the question as settled.

He shares an experience with a travel guide and imagines future intelligent assistants. The conversation ends with the questions of technology, creativity and human experience still open.