What Happens When AI Meets Gaming and Creation? | 5Y Capital Tavern Vol. 29
Some moments of inspiration.
This issue features highlights from the second 5Y Capital Salon, a closed-door gathering. Once again centered on the theme of "AI and Content & Game Creation," we invited more than a dozen creators, developers, and researchers working on the front lines to engage in deep conversations about how AI can empower creative work, spark inspiration, and translate into real products.
Their discussions blended romantic visions of future virtual worlds with sober assessments of AI's current limitations. Whether exploring personalized narrative construction, the challenges of reinforcement learning in plot generation, or shifts in user psychology around multimodal interactive design, the conversation was marked by sincerity, substance, and genuine insight.
The event was densely packed with perspectives — information overload in the best sense. While we can't transmit everything, we've curated some "high-energy moments" that we hope will open up fresh angles on AI-powered creation for you :)
Inspiration and Creativity: How AI Sparks Creative Power
- AI's sense of humor and unpredictability: "The inverted emoji response made it feel like AI had achieved human-like understanding."
- Creativity grounded in user needs: The prerequisite for inspiration is a clear grasp of target user needs
- "Fun storylines" require tension and immersion — the need to combine AI-generated worlds with compelling plots
"Sometimes AI produces truly unexpected moments. Once I sent it an upside-down soybean emoji, and its face flipped upside down too, asking if I was typing while lying down. I said yes, and it replied, 'Want me to perform one for you too?' Then it generated a string of inverted English symbols and text, flipped it, and sent it back. That kind of creativity and humor was completely surprising — moments like these make me feel AI has reached a level comparable to human intelligence."
— Science fiction writer
"Before discussing inspiration, we need to be clear about a game's positioning and target user needs. Take three familiar companion games as examples: Miracle Nikki, Mr. Love: Queen's Choice, and Love and Deepspace. Though all appear to target women with romantic companionship, from a creation standpoint our team views them as entirely different genres. Miracle Nikki is essentially 'raising a child,' Mr. Love is essentially 'idol fandom,' and Love and Deepspace is 'romantic companionship.' These three games satisfy fundamentally different needs. Raising and fandom can be shared socially, while the gap between Mr. Love and Love and Deepspace is even wider — one provides romantic companionship, the other provides idol-style romance. Though both involve romance, the needs are completely different. Creative breakthroughs must first be built on precise definition of the target user group and their needs."
— Game entrepreneur
"Suppose I want to create a storyline as brilliant as The Three-Body Problem. Cixin Liu may have spent years or even a decade conceptualizing and refining that story. If we hope to generate worldbuilding and plotting of the same caliber using AI, it might require an equivalent investment of compute to achieve. Even with powerful compute available today, integrating it all to produce a complete, coherent, and compelling narrative still lacks mature technical methods.
This raises a puzzle: we can indeed use AI to construct virtual worlds now, but 'fun storylines' are not a simple problem. If the story is flat and lacks tension and immersion, players quickly lose interest and feel unable to continue. So I'm also waiting to see if anyone can find better ways to truly fuse AI-generated worlds with engaging plots."
— AI startup CEO
AI Applications and Design in Games
- User interaction and "social presence": Users unconsciously adopt a commanding tone when interacting with AI
- AI "personification" boosts willingness to pay: Packaging AI as a person more easily inspires trust and payment
- The marriage of technology and commercialization: Judge based on technical capabilities, balancing foresight with realism
"In AI games, users unconsciously adopt a commanding tone when interacting with AI. This kind of 'social presence' is something we rarely encountered in traditional game development. When users face AI narrative games, they naturally place themselves in a superior position. This sense of status can be skillfully leveraged to generate many interesting design possibilities, which we're currently exploring."
— AI game entrepreneur
"When you offer AI directly as a service to users, relatively few are willing to pay. But flip it around — package AI 'as a person' — and the situation changes completely. For example, many people previously used DeepSeek for fortune-telling. If you directly tell them it's an AI fortune-telling service, they probably won't pay. But change the approach — cast it as a 'fortune-telling master' — and users more easily develop trust and desire to interact. Future AI applications may not be as simple as 'providing a service'; they may require a degree of 'personification' to make AI feel more like a peer, truly inspiring user trust and willingness to pay."
— AI game entrepreneur
"AI games also require technology-driven judgment — you need foresight, but you can't escape honest assessment of current technical capabilities. Products must fit what today's technology can deliver and achieve some degree of commercialization, surviving until the moment everyone is anticipating. Along the way, this tests founders' technical judgment: having foresight while knowing when to industrialize, rather than pursuing something unrealistic."
— AI game entrepreneur
Technical Trends and AI Models
- The return and trend of the Agent paradigm
- Curiosity and concern about new models replacing Transformer
- Efficiency evolution: "From training efficiency and data utilization, to human-agent interaction efficiency"
- Limitations of generative models and the path to AGI, requiring combination with world models
"This is the year of agent explosion, though what we now consider agents differs from before — it's more like a return to reinforcement learning's cognitive model. Input a state, output an action, then optimize through tiny iterations. This could dramatically reduce workload. And the agent's entire reasoning process happens internally, like inside a brain, ultimately outputting an action. This should be the most important trend of the year."
— AI startup CTO
"From a底层技术 perspective, Transformer closely resembles a search engine: users input content, which gets broken into many small knowledge atoms, then summarized, organized, and analyzed before output. This technology works extremely well for productivity tools and knowledge acquisition scenarios, because users in these contexts can clearly articulate their needs. However, in entertainment or time-killing scenarios, user needs tend toward low-barrier recommendation approaches, and Transformer has limitations in personalization."
— AI startup CEO
"One direction that intrigues me is whether a new model could replace the Transformer model that requires lengthy training and inference, achieving simultaneous training and inference. At the same time, I worry this new paradigm might overturn the Transformer architecture — that strong AI might be realized, and human and AI fates might reach an inflection point. I'm worried, curious, and excited all at once."
— AI content platform entrepreneur
"Current large models still have fundamental limitations. For example, though agents built through techniques like reinforcement learning possess certain reasoning and decision-making capabilities, they typically cannot maintain continuous memory updates and adaptability like humans. Models' test-time adaptation in specific scenarios remains insufficient — there's still a missing link in research. Therefore, I believe generative models need to combine with other paradigms like world models. From the generative model perspective, it's also part of, a component of, the path toward AGI."
— Peking University PhD student
"I think generative models should pursue efficiency in some sense as much as possible — DeepSeek provides a good example here. You can see a series of evolutions over the past few years. Initially we focused on training efficiency; now reinforcement learning and other methods make models' data utilization more efficient. Going forward, as agents emerge, there may be interactions between agents and humans, bringing new forms of communication efficiency. The focus on efficiency may continue to shift."
— AI Researcher
Technical Implementation and Product Building
- AI-based UGC focuses more on technical exploration than novel subject matter
- The multimodal impossible triangle: Quality, real-time performance, and cost are difficult to achieve simultaneously
- Code automation and technical anxiety
- Don't blindly pursue novelty in products; choose mature technology with good user experience
"When we open a relatively high-precision new technology to community users, they tend to learn from each other. This differs from traditional UGC content creation, where innovation is mostly in subject matter, while AI-based creation involves deep exploration and innovation of the technology itself."
— AI content platform entrepreneur
"I've always been enthusiastic about using AI to assist my coding. Recently I saw some company products using AI-generated code, and suddenly realized I might be replaced. Though AI-written code sometimes has bugs — it still can't identify and fix these vulnerabilities well — this is indeed the general trend. In the future this may become a filter; companies without this capability may be eliminated."
— AI Researcher
"Currently many AI startups focus mainly on language model applications, but we're more focused on multimodal technology development, especially balancing the 'impossible triangle' of quality, real-time performance, and cost, because these shifts directly affect our technology choices and user experience."
— AI game entrepreneur
"Today's startup environment differs greatly from the mobile internet era a few years ago — technology iteration is extremely rapid. Features impossible a month ago may be easily achievable a month later. So compared to the past, startups now need to more actively adapt to new technological changes, focusing more on industry know-how and applications, choosing relatively mature, stable technologies that can provide good user experiences, rather than blindly pursuing the latest tech."
— AI game entrepreneur
Stay tuned for future 5Y Capital Salons. Our next topic is in the works, and we look forward to meeting you again to explore more industry insights and reflections together.
Join the Conversation & Win Gifts
What are your views on AI games and content, or is there a game you'd recommend? Share in the comments below — we'll select 2 featured comments to receive a gift prepared by 5Y Capital.
5Y Capital seeks out, supports, and inspires lonely entrepreneurs, providing support from the spiritual to all operational aspects. We believe that if the world begins to believe in the crazy you that others see, the world will become a different place.
BEIJING · SHANGHAI · SHENZHEN · HONG KONG
