100 AI Creators is a weekly series featuring conversations with China’s leading minds in artificial intelligence. As technology evolves, their perspectives shed light on the ideas driving the AI era across borders.
At first glance, it seems like an unhinged social app, but once you’ve tried it, you’ll probably end up recommending it to your friends.
Splat is a social platform built around artificial intelligence, with one unifying theme: verbal combat. As internet communities become more sanitized and polished, Splat plants itself at the edge of expressive chaos, inviting users to argue, vent, and turn emotional warfare into entertainment.
On Splat, users pick sides, form teams, and enter virtual rooms to engage head-to-head in verbal showdowns, complete with matchmaking mechanics and a scoring system.
Splat’s homepage is a feed of polarizing matchups, pitting Mini World or Roblox versus Minecraft, Lionel Messi fans against Cristiano Ronaldo supporters, and Stephen Curry “stans” clashing with LeBron James loyalists. Tap into one, and you’re either assigned to a team, or choose one yourself. Inside, users take turns posting, clapping back, upvoting, and forging alliances.
So how does a team win? For starters, what counts more than the quantity of comments is whether the jibes land strongly. Splat is designed using AI to evaluate and score responses for sharpness, sarcasm, and ability to set the tone.
And sometimes, your opponent isn’t even human. Splat has injected hundreds of AI agents into the mix to stir the pot, amplifying the conflict as digital provocateurs when a push is needed to get things heated.
During its beta test, Splat reported some users got hooked, hopping from topic to topic and trading over 100 barbs a day. One user even befriended a rival midway through a debate, impressed by their ability to throw shade with precision. Others just simply watched, with one user reportedly lurking in a room for 50 minutes straight.
Splat is a haven for keyboard warriors. On most platforms, fighting leads to bans. On Splat, it can earn rewards. The platform embraces a gamified approach to managing the human urge to insult.
Splat was created by Cao Tong, a 26-year-old who graduated from Renmin Business School. He’s a social product experimentalist known for extreme lifestyle stunts like trying to survive on RMB 5 (USD 0.7) a week, and a self-proclaimed addict to all things internet. Cao spends about five hours a day on Douyin, insists he only consumes legal content, and jokes that he’s running out of interesting books to read. His latest purchase? The complete Ultraman anthology.
Cao once quipped that warfare is a primal urge baked into human history, and verbal sparring is simply a low-intensity form of it. “Homo sapiens won out in natural selection thanks to two traits,” he said, “a craving of war and a love of socializing.”
By combining conflict with connection, Splat aims to be a next-gen social platform and one that could replace Snapchat. “We’re building this for Gen Alpha,” he said, referring to kids born after 2010.
Cao has been building products since college or technically, since grade school, when he flipped bottled water for a profit. His first real startup was a campus vending machine company that raised over RMB 1 million (USD 140,000) from angel investors.
But his obsession with social dynamics was sparked during his junior year, when he watched NBA fans go to war online. He realized that arguments created faster bonds than curated conversations. “Celebrity feuds, sports fan brawls—it all looks chaotic. But for young people, that chaos is the shortest path to belonging.”
That led him to build Sandan, his second product. Lacking a technical background, Cao was scammed out of RMB 800,000 (USD 112,000) by shady outsourcing vendors, churned through three dev teams, and eventually shut the project down, burning through RMB 10 million (USD 1.4 million) in the process.
He came away from that experience with three rules:
- The fastest way to validate demand is ad spend. If paid ads don’t attract users or are too expensive, the demand isn’t there. Don’t lie to yourself. Kill the product.
- What users think, say, and do are three separate things. Don’t waste time on user interviews. Ship fast, then study user behavior and metrics. That’s the real research.
- Never hire unless absolutely necessary. “I’m done collecting bosses,” he said, referring to past hires that became liabilities.
Splat is his third attempt and once again, it’s built around online fighting. Cao believes there’s still an unmet need in the internet age: the desire to argue. “Why keep suppressing it? If someone doesn’t get it, I’m not here to explain. Fighting will be the bedrock of next-gen social networking.”
The following transcript has been edited and consolidated for brevity and clarity.
Stoking the fire using AI
AI Now! (AN): Splat continues the “verbal combat” concept from your past work. What’s new?
Cao Tong (CT): Our earlier products had three core issues.
The first was a low matchmaking success rate. Think of Honor of Kings. If you fail to queue several times in a row, you’ll probably stop playing. Our previous average matchmaking rate was just 18%. That’s essentially a death sentence for any social product.
Now with Splat, we’ve introduced AI agents. The goal is for them to behave like real users, not just NPCs parroting preset lines, but digital personas that can observe, choose sides, talk trash, and exit like real people. Once, I dropped 200 agents into the system, and engagement spiked. Retention also visibly improved. If someone’s ready to throw hands, the agent’s job is to light the fire and build a social atmosphere.
The second issue was cheating. In our old app, Sandan, some users spammed copy-paste lines or used third-party keyboards to flood the chat. We tried rules-based detection, but cheaters always found a workaround. In any game, anti-cheating is a serious challenge. H1Z1, one of the earliest battle royale games, saw its player base collapse largely due to poor anti-cheat enforcement.
Still, rampant cheating can also be a good sign: it shows users are invested. Nobody goes through that much trouble for a boring product. With Splat, we now use AI to detect cheating. We feed the model valid sample messages and examples of how past users cheated so it can flag bad actors. The core goal is to filter out spam so users can focus on quality trash talk.
The third issue was judging content quality. We used to reward points based on word count or number of likes, but that doesn’t necessarily reflect whether a roast hits. Sometimes a single meme can dismantle someone better than 800 words of nonsense. Now we tap AI to identify what counts as a high-quality insult and assign better scores accordingly.
AN: How does AI score good trash talk?
CT: Two metrics. First, emotional response: does it provoke a reply? Second, structural awareness: is there satire, timing, humor? Splat rewards smart insults, not profanity-laced rants.
AN: Other platforms use AI agents too. What makes yours different?
CT: We analyzed heated arguments across platforms worldwide and found that online fights are surprisingly universal across cultures. But the behavior behind them is more nuanced than most people think.
Some users just want to watch. Others drop one line and disappear. Some hop between sides. Some go undercover in the enemy camp. So we designed agents with a range of “personality models” to accommodate various styles.
The agents don’t follow scripts. They decide how to engage based on topic popularity, team balance, and the flow of conversation. They adjust tone and intensity in real time, depending on user reactions. Beyond recognition and mimicry, we’re training them to spark emotion. If the rhythm stalls, an agent might shift the topic to reignite the exchange.
We clearly label agents as AI. There’s no deception. The goal is still to build real human connections.
AN: What large models do you use?
CT: We use whatever is affordable. We’ve used Qwen and GPT. We skipped DeepSeek as it sounds too cultured for what we are doing.
AN: Could Splat work without AI?
CT: AI is a flywheel, not a feature. It solves three key problems for us: better matchmaking, blocking cheaters, and ranking good arguments.
The real challenge in building a product is getting people to actually use it. The first Tesla vehicles were buggy, but once they shipped, they improved fast.
With social apps, the question isn’t how to add AI, but understanding what makes people stay. Social products must be fun and simple, not obsessed with AI-driven features. We serve users, not the AI hype machine.
AN: Do people accuse you of slapping on AI for the hype?
CT: We rely on AI more than it sounds. All three of the problems I mentioned are solved with AI, and it has made a big difference.
What I can’t stand are founders who slap AI on everything without knowing what they are actually building. They start with a vague idea and pitch it as a solution using AI.
Splat is a real AI product. We use it to solve specific, clearly defined problems. Just because it sounds technical and not flashy doesn’t mean it hasn’t incorporated AI enough.
In the AI era, I’ll never swing a hammer just to look for nails.
Connecting through verbal combat
AN: You’ve said that 90% of Gen Z will eventually use Splat. Why do you think fighting is the new social entry point?
CT: Wang Xiaobo once said that life is just a slow process of being worn down. Adults take the hits and eventually grow numb. But teenagers still carry a deep urge to connect with the world.
Here’s a typical use case: a 17-year-old Kobe Bryant fan is scrolling online when he suddenly sees a post that reads, “LeBron stans are better, Kobe fans are trash. I’m waiting on Splat. Come fight me.” He’s triggered, grabs a few friends, and jumps into a room. They trade insults, go back and forth, then maybe laugh and add each other as friends afterward.
Fighting taps into two things: self-recognition and group belonging. Most people struggle with high-quality expressions. They can’t articulate their thoughts or get feedback. But choosing sides and yelling? That’s easy. That’s universal. It’s why tribal clashes have always been part of internet culture.
AN: Doesn’t it feel “wrong” to build a product that encourages arguments?
CT: Only if you lack empathy. I view users with respect. I hate moral superiority, the kind that comes from judging others from a narrow ethical pedestal.
This isn’t about inciting violence. It’s about channeling one’s expression. Think of it as a relief valve. Just as repression breeds dysfunction, open expression fosters understanding. Splat is structured chaos. And I stick to neutral topics like games and sports, nothing political.
AN: You’ve worked on similar apps before. Do you have proof there’s demand?
CT: That’s the fun part. If we asked someone on the street whether they would use an app to roast people, they would probably say that isn’t really their thing. But then they would sign up for Splat. On other platforms they post about peace and positivity, then come to Splat and go absolutely wild.
What people think, what they say, and what they do are three separate things.
Back when we were testing this in China, we ran off-platform ads and got incredible ROI. For every 1,000 impressions, we had 2,000 comments. That was a clear signal. The demand is real, and the engagement is intense. The users we brought in were 60% male, 40% female. So no, it’s not just guys yelling on the internet.
People used to rally around bloodlines, religion, class. Now it’s interests and identity. We turned that into a game. Let people release tension online, and they are less likely to carry it into the real world.
AN: Don’t other platforms already host plenty of arguments? Why should people download yours?
CT: Look at Weibo, Zhihu, Xiaohongshu. People argue there, but they are not built for it. That’s like hosting a house party with no toilet. If you take a dump in the living room, of course people get mad.
We built a mansion with a proper bathroom. You can rage in peace. Your emotions get processed. And sometimes you even make a new friend.
We’re making argumentation safe and structured.
AN: What about when fights get out of hand? Ever worry about people taking it offline?
CT: Honestly? We’ve never seen that happen. Whatever someone needs to say, they say it right there in the fight. That emotional pressure gets released. In real life, they are calm.
Splat is like a hippie sandbox. The more you argue here, the less serious it feels. Emotions get gamified and once they are part of a game, they lose their sting.
AN: How do you handle content risk?
CT: I don’t think we face more risk than anyone else. We’re using a gamified format to help people release emotion. And we only focus on general topics like games and sports. We don’t touch anything beyond that.
Wasting time is part of the plan
AN: Is Splat a reflection of your personal values?
CT: Definitely. My first principle is: nothing is off limits, as long as it doesn’t hurt others. If a need exists, it deserves to be met. Venting is a need both for individual and group identity. It’s a way for people to understand themselves better and form stronger connections with the world.
Another belief I hold is that suppression should be solved by release, not restriction. Look at the past when sexual repression was intense, more twisted behaviors surfaced. Now, things that once seemed strange are as ordinary as eating. There’s less harm when things are normalized. The same goes for gaming. Many kids who get addicted to games are being heavily restricted by their parents. I loved games as a kid too, but I never got addicted because no one stopped me. If someone told me to study, I’d just pause the game and go study. A lot of parents lock phones away, but that’s not the answer. Society needs to let pressure flow, not build it up.
AN: Do you personally enjoy arguing online?
CT: Not really. I’ve never been the angry type. But I’ve spent hours arguing on Splat using a fake identity. I’ve made friends that way. Some users even created memes of me: black eyes, swollen faces. I love it.
My dad jokes that I’m still mooching off him. I say, sure, and remind him Liu Bang, the first emperor of the Han dynasty, got the same from his dad.
AN: What’s the biggest criticism you face right now?
CT: That the product lacks values. Most investors come from an elitist mindset. They can’t fathom that fighting is a real, widespread need. I’ve been immersed in this product for a long time. I argue with users every day. No one understands this niche better than I do.
The problem today isn’t too much yelling. It’s that people have too little space to let it out.
I don’t believe in viewing the world through an elitist lens. That kind of thinking flattens people and erases their complexity. Zhang Yiming once said he wanted to be morally perfect. I think that’s silly. I loved Neihan Duanzi, that app was hilarious. I was reading it in eighth grade.
To be morally correct just means forcing uniform values on messy, complex people.
AN: Some say users are just wasting time on Splat. Is that really the future you want?
CT: Let me ask you this: are there more people trying to save time, or kill time?
We always say tech should help us save time. But what do people do with the time they save? They waste it anyway.
A lot of human progress comes from wasting time. I created Splat because so much of my own life has been “wasted.” That’s why I understand what people want from entertainment products. This industry isn’t about ROI or productivity. It’s about passing time.
I’m the poster child for that. I was playing League of Legends at an internet cafe the night before my college entrance exam.
AN: Do you still waste time every day now that you’re running a company?
CT: More than ever, but now it’s called work. Scrolling Douyin is my job. It’s my version of reading the great works.
AN: What’s the most important metric for a next-gen social app?
CT: Long-term retention. And that means building lasting relationships.
On most social apps, it might take ten actions to become friends. On Splat, it takes three.
Snapchat’s genius wasn’t disappearing messages. It was capturing the feeling of passing notes in class. You could talk to a stranger face-to-face ten times and still feel like strangers. But pass just three notes in class, even if they are filled with nonsense, and suddenly, a connection forms.
Fighting works the same way. It’s one of the fastest, most honest ways I’ve seen to break the ice and build bonds.
RELATED ARTICLE
Quick takes
AN: What’s the most shocking AI development you’ve seen in 2025?
CT: Tesla’s progress on humanoid robots and self-driving. We’re getting dangerously close to the point where machines can replace humans entirely.
AN: What do you think is overhyped—and most underrated—in today’s AI landscape?
CT: Productivity tools are overhyped. Only 1% of people truly optimize for efficiency, yet 1,000 tools are built for them. Most people are far lazier than they admit.
Underrated? Consumer-facing products, especially those meant purely for entertainment. That’s where the real mass market lies.
AN: Who’s the entrepreneur that has influenced you most?
CT: Elon Musk. He takes hit after hit and never falls.
The other day, I was eating at McDonald’s when Tesla’s stock had just dropped. A group nearby started trashing Musk, questioning why he always puts himself in risky positions. One of them said, “A wise man avoids the edge of the cliff.” I didn’t bother responding. People like that will never understand.
If you want to reshape the world in a meaningful way, you can’t be afraid to get your feet dirty. If losing USD 1 trillion brought the world one step closer to his vision, Musk would do it without hesitation.
I’m that kind of person too. I started my first venture in college. I never cared about grades, never thought about money, or what would happen if I failed. Even now, I’ve never paid myself a salary. During the hardest times, I poured everything I made trading crypto back into the project.
I work out of a free coworking space. At lunch, I check to see if other teams left any boxed meals, and I take them.
When you truly want to build something, you don’t weigh the pros and cons. You just go. If the world is run by people who always choose the safest option, it’ll never move forward.
AN: What are three books you’d recommend?
CT: My favorite author is Wang Xiaobo. I recommend starting with “The Silver Age.”
Second is “Zizhi Tongjian,” a study of Chinese history that helps distinguish enduring values from outdated or harmful ideas.
Third is “The Story of Romans” (published by CITIC Press Corporation). It’s incredibly fun and readable, and that’s important to me. If a book isn’t readable, I won’t finish it.
100 AI Creators is a collaborative project between AI Now! and KrASIA, highlighting trailblazers in AI. Know an AI talent we should feature? Reach out to us.