Unique Research · AI Industry Watch · 2026-10-12, Shanghai
95% of AI apps never take off. The time-saving business has been cornered by the big platforms.
“Twelve months to reach $10 million ARR, or we walk away. It doesn’t matter whether it’s China or the US.”
That’s Zhu Xiaohu talking. During TOKEN2049 in Singapore, Google Cloud, Master Concept, and Unique Research put together a “Chinese Founders’ Night” — and the token on the table wasn’t the one being hawked outside the venue, but the token of large language models.
Two men on stage: Zhu Xiaohu, managing partner of GSR Ventures, and Jason Liang, general manager of Master Concept. One manages capital, the other manages the ecosystem.
The moderator was Wu Wei, founder of Unique Research. He opened by ribbing Zhu about his profile picture — still the Bored Ape. “Are you on the Web3 side or the AI side?”
Zhu’s answer was short: “Web3 barely has any opportunities left either.”
Fine. Let’s talk AI.
Why hold this event in Singapore? Wu Wei gave the background: Unique Research and Master Concept started holding events in Singapore two years ago. His read: Singapore may well be the first stop for Chinese founders going global with AI.
Back in March, Zhu Xiaohu attended our event in Shenzhen, when everyone was talking about OpenClaw. Six months later, that hype has cooled fast.
Trends come and go, but the coordinates of these two men don’t change: one reads investments through data, the other reads the ecosystem through what actually gets built.
In the first half of the year, a popular line made the rounds: there’s nothing left to invest in among AI applications.
Zhu says that when he said this himself in H1, he was a bit less sure — his AI app portfolio wasn’t large enough yet. Now he can say it with confidence: AI applications have genuinely exploded. One company he invested in has been doubling revenue every two months since last year. Another — a global, consumer-native AI app company — can reach Honor of Kings-level revenue in two years.
But he immediately offered a brutal symmetry:
“With AI today, you either get the speed right and take off, or you don’t take off at all. 95% of them never take off.”
What if you don’t take off? Wu Wei asked.
“Then it’s on you — the PMF is wrong, the go-to-market is wrong,” Zhu said. The core thing is to change, but sometimes you genuinely don’t know how. If you can’t change, can’t reach that speed, then AI entrepreneurship loses its meaning.
That speed has a quantified standard: early on, 20% week-over-week growth. For early-stage deals, they look at weekly numbers.
Twelve months without hitting $10 million ARR — give up.
You might ask: isn’t that bar too high? Actually, read it the other way. When the bar is low, investors listen to stories. When the bar is high, investors only look at numbers.
The numbers era is actually good for founders: clear pricing, no need to guess whether the investor is in a good mood today.
This topic wasn’t pulled from thin air. Wu Wei said he’d just seen a news item that day: a number of executives were leaving the embodied-intelligence track, with some worrying that robots would be built and nobody would use them.
At the start of the year everyone was chasing OpenClaw; now it’s personal agents. Instinct, for example, reached a $10 billion valuation with just a few dozen people.
Trends get more elusive one after another. Is there a bubble, and where are the opportunities?
Zhu named two places.
The first: the mass of Personal Agent projects in the US. “Pure bubble.” His reasoning is blunt: China did this a decade ago — the Laiye Technology crowd. Back then AI wasn’t mature; things were called agents, but there were real humans behind them.
And the scenarios all ended up flowing into ride-hailing and food delivery, because those are the only high-frequency scenarios.
Now some US agent projects (like Meta’s newly launched Muse) do 50% travel. What’s the frequency of travel? An ordinary person travels two or three times a year at most.
To push DAU up in a low-frequency scenario, there’s only one road left: subsidies. And that’s full circle back to the old playbook from ten years ago.
The second thing to avoid is everything that saves time. For AI built to save time and boost efficiency, Zhu’s exact words: “there’s rarely an opportunity for startups,” because the big platforms will do it themselves.
His big judgment: today’s AI competition is a China-versus-US competition. He said it on his way to Singapore.
In that landscape, save-time tools get made by the giants almost as an afterthought. Startups squeezing in are just running alongside the giants as extras.
What he likes is the opposite direction: kill time — occupying the user’s idle hours.
Ordinary users have plenty of idle time. That’s a market the save-time logic can’t hold.
The example is short drama. Now 90% of short dramas are AI-generated. Many of his friends are glued to Hongguo short dramas. He himself is too busy to watch, but his experience with long-form series is fast-forwarding; if it’s not good, he drops it.
This track is beneath the big platforms’ notice. ByteDance succeeded, and in the nooks and crannies that remain, there’s still opportunity.
One sentence sums up the judgment:
“The giants have cornered the time-saving business. It’s the time-killing business that’s left for founders.”
Opportunity is in kill time, not save time — because most people’s time is surplus.
There’s an even more widespread anxiety: will large models eat the entire application layer?
“LLMs can’t eat it,” Zhu said. Right now the model companies can’t even handle their own price wars — there’s no difference between model companies; it’s just whoever prices lower.
His next observation was more specific. As models advance, they over-optimize in certain professional domains. GPT is great at math and physics problems, but it’s lost the ability to chat.
His metaphor: the model is like a programmer — excellent at coding, excellent at math, but a bit “wooden,” no longer chatting like the earlier models did.
So it’s not that the newest model is best, nor the most expensive — it’s using the most appropriate model for your scenario.
That’s what good companies all do, and the model companies themselves have no idea.
Look at GrokBot. Musk said you can use Opus for free; he doesn’t even necessarily push his own model — you can use others, even Gemini inside.
This line is worth writing down:
“The model is the shelf; the application is the chef.”
Diners only care whether the dish tastes good, not which brand of soy sauce you used.
Switch to Jason Liang. Master Concept is a top Google Cloud partner, having spent years accompanying wave after wave of Chinese teams overseas on deployment and commercialization. He talks about the pitfalls.
First, the strength. Jason believes the biggest advantage Chinese teams have going global is one character: hustle.
“Chinese founders are relentless. We really work hard — taking the time others use for leisure and putting it into work, iterating endlessly.”
Bringing what’s been proven in China overseas, doing it faster, matching local tastes better — nothing stops you from doing that.
What stops you are the pitfalls. The biggest is hidden cost. Customer acquisition cost, infra cost — those everyone can calculate.
What you can’t calculate is security and compliance cost. GDPR and the like. Early on you won’t put it on the books as a cost, because it’s unforeseeable.
It runs fine in China; you go overseas, and some things become non-compliant. Some data collection is illegal without local storage. Then the fines come.
He cited Plaud and Even Realities. Their founders came from companies like Apple, so early on they walked the entire bottom-layer architecture and full chain through security and compliance.
Many domestic founders, used to Alibaba Cloud, domestic platforms, or self-built servers, assume they can move overseas by simple transfer — that’s the easiest place to step on a landmine.
So when you first go overseas, where should the first money go?
Jason’s answer: not a cent on infra. Google has a startup program — all the big companies do. Harvest the subsidy first.
The subsidy is phased: submit a website at the early stage and you get $2,500; later it goes to $25,000; with a Pre-A round and credible investor backing, $100,000 and $250,000.
Besides cloud resources, you can also harvest booths and endorsements. At their events in Singapore and the US, they’ve put Plaud and Even Realities on stage — product built on Google’s architecture, security and compliance you can trust.
The one thing you genuinely should spend money on is cheap: register your company’s account system properly from day one.
Don’t just sign up with a personal Gmail and start working. If one day your phone number changes country, your IP changes, the account is gone — nobody can get it back for you.
Gmail has billions of users worldwide and is run by a dedicated team; even the big companies make this mistake. What does it cost? A few dollars a month.
Wu Wei testified on the spot: after Jason’s last reminder, he went home and registered right away, because he also runs business with overseas users. Even the event organizer was the one being enlightened.
Sitting in the audience, what I took away: the first lesson of going global isn’t how to make money — it’s how not to lock yourself out of the door.
Many founders agonize: fundraising drains energy, but without money you’re short on funds — which comes first?
Zhu splits founders into two types. One is the star founder — big-platform halo, top-university PhD, a company that tells stories. The other runs the company by the numbers. He prefers the latter.
“If you can grow 20% week-over-week, you’ll definitely have money. If you can’t hit that speed, early fundraising is useless anyway.”
He’s seen plenty of companies that raised money on PPTs, backgrounds, and résumés over the past two years. Two or three years on, they have nothing. The product doesn’t dare launch — too much baggage. The moment it launches, the numbers don’t match, and the company is “finished.”
Wu Wei chimed in: “So the first round is hard, but the follow-on is harder. If it were me, I’d never launch the product — too big to fail.”
That was, of course, a joke — but it hits a real problem: not daring to launch is, at bottom, knowing the data can’t stand scrutiny.
Jason added a more operational cut: “You don’t even have a business yet — why do compliance, why build core architecture? Just put it on Google Cloud.”
Take the giants’ credit, let your engineer co-founder attack the core business value. Once the value shows, investors come naturally, and the money goes to the right places.
Put plainly, the combination is: the giants supply the foundation, the founder supplies the value, and investors supply the accelerator. The order must not be reversed — reversed, you’re just burning money on PPTs.
Finally, on advice for Chinese founders, Zhu offered some views:
“Today I think going global is mandatory. Whether to C or to B, the commercial potential of AI applications in overseas markets is many times larger than in China.”
He compared two generations of founders. Back then, Didi, Meituan, and JD all had trouble internationalizing — that generation of internet founders was relatively conservative.
Today’s founders are used to internationalization. It works far better, and commercial value is far easier to create.
Then two numbers.
First: the most successful project he’s backed took its first check of just $1 million. After that he was basically chasing the founder to take more money, because the speed was there and it wasn’t burning cash. The company is now very large, yet has only raised about $20 million in total, most of it still sitting on the balance sheet.
His conclusion: starting a company today genuinely doesn’t require money. “If you need to burn a lot of money, you’re doing it wrong.”
Second: an open-source project they backed, Open Design, hit 100,000 GitHub stars in four months. The company has just 20 people, but more than 500 organizations worldwide help it develop, acquire users, and build reputation — and only then convert to purchases. That’s the meaning of the open-source ecosystem. And it’s the right playbook for an ordinary founder building AI apps.
These two numbers read well together. A $1 million seed and 20 people were shabby by mobile-internet-era standards; today they’re lean.
The leverage in the AI era isn’t in how much money you have — it’s in speed and ecosystem.
My own read: this round of going global is not the same as the previous internet generation’s.
The last generation copied China’s model outward — at bottom, an export. This generation is born from day one on a global foundation — global clouds, global open-source communities, global compliance frameworks. The company may still live in a residential building in Shenzhen or Hangzhou, but the business is already a world business.
The threshold has come down, yet the standard has gone up. That 20% weekly growth rate is the new common language.
Jason’s advice was plainer, a clear path for early teams: “Apply for the giants’ credits, use the giants’ security foundation, borrow the giants’ endorsement for marketing, and then go win your own kingdom.”
One sets the standard, the other gives the path. This panel was really the elaboration of those two sentences.
Before wrapping up, Wu Wei asked: what AI tool do you use most often now?
“Doubao, of course.”
Zhu said his key use is making PPTs. Voice input matters a lot — it can recognize various dialects and mixed Chinese-English speech.
Doubao isn’t the best model, “but for 99.99% of users, you don’t need the best model.”
His questions are all basic — like which Chinese restaurant in Singapore is good lately. Does that need the best model? No.
Jason pressed: “And paying?”
“I certainly don’t pay. I have no need to pay.” He made a PPT with Doubao; halfway through it warned him the quota was used up and he’d need to pay. Reset, and you get another free PPT. When he used that up, he switched to WorkBuddy — also good, and free.
Jason seized the opening: “Next time you’re overseas, try our Gemini — our overseas data is pretty good.”
Zhu’s focus was elsewhere entirely; he uses Gemini to generate charts, then puts them into his PPTs.
This little sketch is the best footnote to the whole conversation. An investor who watches weekly growth and welds the ARR bar at $10 million keeps his own AI standard almost comically simple: as long as it understands me, and it’s free.
Model companies compete on parameters; investors compete on numbers; and users just want to know which Chinese restaurant in Singapore is good.
Figure out who creates value for whom, and you’ve probably found the door handle to this wave of AI going global.
Host: Wu Wei, founder of Unique Research
Guests: Zhu Xiaohu, managing partner of GSR Ventures; Jason Liang, general manager of Master Concept
Wu Wei: Everyone knows Jason. I believe everyone knows Professor Zhu too. Let me warm up with a few questions — I see your profile picture is still the Bored Ape. Between Web3 and this AI wave, which do you think has the bigger opportunity? Which side are you on? At least the avatar suggests you’re on the Web3 side.
Zhu Xiaohu: Web3 barely has any opportunities left. The crypto crowd is trading US stocks now — that shows crypto people themselves don’t believe in digital currencies anymore.
Wu Wei: OK, back to the topic. The token we’re talking about today isn’t TOKEN2049’s token — it’s the token of large language models. Today we’re mainly talking about Chinese founders going global, AI, and the token economy. Your perspectives are very different — one from capital, one from the ecosystem.
Wu Wei: Let me start with you, Zhu. In March we held an event in Shenzhen when everyone was talking about OpenClaw — nobody expected the hype to cool so fast within half a year. You’ve said we can’t just talk narrative — no more storytelling — you have to look at ROI. You’ve always stood on the AI application side and think this wave has a big opportunity. Have your investment standards changed compared with before? There are several AI founders on stage today.
Zhu Xiaohu: The bar is higher now than before. In the past two years I’ve looked at too many AI apps. As long as you do it right, the speed can be extraordinary — we have a company that has been doubling every two months since last year. Today with AI, you either get the speed right and take off, or you don’t take off at all — maybe 95% never take off.
Wu Wei: What if they don’t take off?
Zhu Xiaohu: Then it’s on you — PMF is wrong, go-to-market is wrong. The core is to change, but sometimes you genuinely don’t know how. But if you don’t take off and can’t reach that speed, then AI entrepreneurship is meaningless.
Wu Wei: Is there a benchmark for this speed? Like ARR?
Zhu Xiaohu: Twelve months without hitting $10 million ARR — give up. Whether in China or the US, it’s the same. Early on, it’s 20% week-over-week growth. Data like that has the chance to explode quickly. For early-stage deals, we generally look at weekly data.
Wu Wei: Weekly growth. Back in Shenzhen, everyone was building around the agent ecosystem. Today we’re in Singapore — you asked how we started doing events here. Because Unique Research and Master Concept started events in Singapore two years ago. I’m very bullish on the Chinese-founder AI globalization opportunity — Singapore may be the first stop for globalization. Now some voices say there’s nothing to invest in among AI apps. What do you think? Where’s the opportunity?
Zhu Xiaohu: In H1, saying that made me a bit unsure — our AI app portfolio wasn’t large enough, not enough confidence. But now I can proudly say AI applications have truly exploded, very fast. We have a company that can reach Honor of Kings-level revenue in two years — this is very obvious with global consumer-native AI app companies. It fully shows this isn’t a bubble; there’s a big opportunity.
Wu Wei: So you still firmly stand on the application side — you don’t think the big models will eat everything?
Zhu Xiaohu: LLMs can’t eat it. The model companies can’t even handle their own price wars right now. What difference is there between model companies? No difference at all — it’s just whoever prices lower.
And we’ve found something interesting: as models advance, they may over-optimize in certain professional domains. For instance, GPT is now very good at math and physics, but it can’t chat anymore — just like a person: the model is like a programmer, excellent at coding, excellent at math, but it’s become a bit “wooden,” no longer chatting, not like the earlier models.
So AI really is like humans — it’s not that the newest model is best or the most expensive is best. You have to use the most appropriate model for your scenario. That’s what the good companies do — and that’s something the model companies have no idea about.
Look at GrokBot. Musk says you can use Opus for free; he doesn’t even necessarily push his own model — you can use other models, even Gemini’s.
Wu Wei: That’s the capital angle. Jason, you represent the ecosystem angle. Over the years you’ve accompanied many teams on deployment and global commercialization, and you’ve seen many cases. What’s the most important lesson, and what’s the biggest bottleneck?
Jason: The most important lesson, which I mentioned at the start — Chinese founders are relentless. We really work hard, taking time others use for leisure and putting it into work, iterating endlessly. Just as Zhu keeps saying — be fast, fast in every way. Take what’s good about us overseas, leverage our strengths, do it faster, match local tastes better, create more value.
But today I want to talk more about bottlenecks. The bottleneck is that some hidden costs are sitting right there. Customer acquisition and advertising costs, bottom-layer infra costs — these everyone can predict. But I think one big hidden cost everyone must avoid early is security and compliance cost.
The problem with compliance cost: early on you don’t treat it as a cost, because it’s unforeseeable. Like GDPR. Once you’re running well domestically and take things overseas, some things become non-compliant; some things are illegal without local data collection. These fine-related costs, I think everyone needs to watch out for.
Early teams like Plaud and Even Realities — their founders already had relevant knowledge and experience, maybe having worked at Apple and similar places, so they were thorough on this — early on they checked all the bottom-layer architecture, the whole chain, whether the model was security- and compliance-compliant.
Because many domestic founders are used to Alibaba Cloud, domestic platforms, or self-built servers. But when going overseas, you can’t take that architecture for granted — you still need to look at each country’s and region’s laws and policies.
Wu Wei: So not everything needs to be done in-house; you can hand things to ecosystem partners. Where should the first money go when you first go overseas? Email?
Jason: The very first step going overseas, the first money — don’t spend it on infra and bottom-layer architecture. Besides Google’s startup program, all the big companies have programs. Use those programs first, harvest the subsidy.
Besides subsidy harvesting, what else? For example, when we hold events on Google, like Plaud and Even Realities — the reason we could present them: we often give them booths for overseas promotion, saying “my product is this great, built on Google’s architecture, safe and compliant, use it with confidence.” At our events in Singapore and the US, we’ve given Plaud and Even Realities plenty of these scenes. That’s also something you can harvest.
But the thing you actually need to spend money on is very little — one foundational item, like Wu mentioned: the account. If you used a personal Gmail or some casual registration, and one day your phone number moves to a different country or region, your IP changes — it’s gone. Even if you talk to us, or talk to Google colleagues, you can’t get it back. Gmail is used by billions of people and is run by a dedicated team; even the big companies make this mistake.
I hope Chinese companies that are just going global and just starting to globalize will secure the bottom-layer account stuff well. It doesn’t cost much — a few dollars a month.
Wu Wei: Right. Last time Jason reminded me, I went and registered right away, because I also have some overseas-user business now.
Wu Wei: Next topic. One of you represents capital, one the ecosystem — both can weigh in. Many team founders are torn — they need money to develop, but fundraising is hard, and spending time fundraising means less energy for the business. Raise first, or run the business first? Build capability yourself, or lean on ecosystem partners? Zhu first?
Zhu Xiaohu: There are two kinds of founders now. One is the star founder — big-platform halo, top-university PhD — a company that tells stories. The other runs the company by the numbers. We prefer the ones who run it by the numbers. Basically, I don’t think you need to match with investors; it’s meaningless. If you can grow 20% week-over-week, you’ll definitely have money. If you can’t reach that scale, that speed, early fundraising is useless — honestly. Or you really can tell stories — but telling PPTs is honestly harder now.
In the past two years, a lot of people raised money on PPTs, backgrounds, and résumés. Two or three years on, they have nothing, the product hasn’t launched. After two or three years it’s over, and then the product doesn’t dare launch — heavy baggage; the moment it launches the company is “finished,” none of the metrics match.
We instead like those who speak with data and carry no baggage — who can genuinely keep changing and keep probing for the direction. Those can actually take off, and fast.
Wu Wei: So the first round is hard, but the follow-on is harder. If it were me, I’d never launch the product — too big to fail.
Wu Wei: Zhu, there are many opportunities and directions, some narrative-driven. Today I was chatting with colleagues — it seems there are opportunities we haven’t entered, like embodied intelligence and AI hardware; we’ve stayed with AI applications. From OpenClaw at the start of the year to personal agents — a company called Instinct is already worth $10 billion with just a few dozen people. Which tracks do you think have the biggest bubbles? I read news today that many embodied-intelligence founders are leaving — Guo Lie left, worried the robots will be built but nobody uses them. Where are the bubbles to avoid? Which have better future commercialization opportunities?
Zhu Xiaohu: I think it’s the same everywhere now — the US bubble is bigger; China has bubbles too. The mass of Personal Agent projects in the US is pure bubble. China did this 10 years ago, in the Laiye Technology era — even though AI wasn’t mature then, they’d already tried all sorts of AI agents. The scenarios all ended up subsidizing ride-hailing and food delivery. Because honestly, even if you call it AI, back then there may have been humans behind it — not AI, real people. But also there was no scenario. The high-frequency scenarios are ride-hailing and food delivery.
Look at some US agent projects now (like Meta’s newly launched Muse) — 50% are travel. Travel is a very low-frequency scenario; an ordinary person travels two or three times a year at most. Priority one is you can’t grow anymore; to really push DAU up, you’d have to subsidize. What’s the point of subsidizing it up? It goes back to the old path.
So this morning on my way to Singapore I said: today’s AI competition is China versus the US. For AI that saves time and improves work efficiency (save time), there’s rarely an opportunity for startups — the big platforms do it themselves. For startup companies, the US might still have some vertical opportunities.
But I prefer kill time — occupying the user’s idle time.
Wu Wei: AI short dramas.
Zhu Xiaohu: Right, short dramas. Look — 90% of short dramas are now AI-generated, and the quality is good. Many of my friends are watching Hongguo short dramas — AI-generated Hongguo, watching with relish. I’m busy and haven’t watched yet, but with long-form series I feel there’s nothing good to watch — all fast-forward, and if it’s not good I drop it. Short drama is a real opportunity. Helping users kill time — I think that’s a unique company opportunity. The big platforms don’t care about this now — especially ByteDance. ByteDance is very steady; it succeeded. In the remaining nooks and crannies there are still opportunities it overlooks. Startups have some room here.
Wu Wei: Opportunity is in kill time, not save time — ordinary users have plenty of surplus time.
Wu Wei: Jason, on money versus ecosystem, how should founders choose?
Jason: I very much agree with Zhu — if you don’t have a product and go raise, even if you close a seed or Pre-A, you can’t move forward. We’ve seen too many of these companies. Early on, working on your own project, investors can give money, but you can also apply for the big companies’ credits — I’ve said this several times. With that credit, the bottom-layer architecture is done; you have your own idea; outsourcing is so strong now; you have your engineer co-founder — so you go hard at your core business value. Once you bring out the core business value, you can attract investors, then put money to work where it counts.
Early on, I think some founders really overthink — building their own core architecture, doing security and compliance. I say: you don’t even have a business yet — why compliance, why core architecture? Just put it on Google Cloud. Early on, they won’t even give you money for that.
Google Cloud’s startup credit isn’t handed out casually, but it’s phased: at the very early stage, submit a website and you get $2,500; later, $25,000; then with a Pre-A and good investor backing, $100,000 and $250,000. Each stage has its focus. Early on, there’s genuinely no need to raise with a PPT before you have a product — that’s my view.
Wu Wei: What people in the room are hearing today is the most authentic first-hand material — this is what AI can’t replace. Can you two share reminders or advice for Chinese founders who are already globalizing or planning to? Jason first.
Jason: My advice is simple — especially at the very start of entrepreneurship: harvest the big companies’ subsidies, use the big companies’ security foundation, borrow the big companies’ endorsement for marketing, and then go win your own kingdom properly.
Wu Wei: Zhu, any advice?
Zhu Xiaohu: Today I think going global is mandatory. Whether to C or to B, the commercial potential of AI applications in overseas markets is many, many times larger than in China. The ceiling domestically is fairly limited now. And going global matters — innovative founders are braver about internationalization than before. Back then Didi, Meituan, and JD all had the problem of internationalizing late — that generation of internet founders was relatively conservative. Today’s founders are used to internationalization. It works far, far better today, and commercial value is easy to create.
Also: starting a company today genuinely doesn’t need money. If you burn a lot of money, you’re doing it wrong. One project we backed very successfully took its first check of $1 million; after that I was basically chasing him to take money — the speed was extraordinary and it didn’t burn cash. It’s now very large, and it only raised about $20 million, most of it still sitting on the balance sheet.
There’s also an open-source project we backed, Open Design — 100,000 GitHub stars in four months. The company has just 20 people, but more than 500 organizations worldwide help it develop, acquire users, and build reputation, and only then purchase — that’s the meaning of the open-source ecosystem. And it’s the right playbook for an ordinary founder building AI apps.
Wu Wei: I’m curious — what AI tool do you use most often?
Zhu Xiaohu: Doubao, of course.
Wu Wei: You seemed to mention WorkBuddy too.
Zhu Xiaohu: My key use is making PPTs. I open it regularly — voice input matters a lot: it recognizes various dialects, and mixed English-Chinese. That’s very important.
Doubao certainly isn’t the best model, but for 99.99% of users, you don’t need the best model. What matters is that it understands me and gets what I say in. The questions I ask are all basic — like which Chinese restaurant in Singapore is good lately? Does that need the best model? No.
Jason: And paying?
Zhu Xiaohu: I certainly don’t pay. I have no need to pay at all. I had Doubao make a PPT; it reminded me the quota was up and I’d need to pay, asked me to reset — reset, and you can try another free PPT. When I used it up I switched to WorkBuddy — also good, and free.
Jason: Next time you’re overseas, try our Gemini — our overseas data is pretty good.
Zhu Xiaohu: I mainly use Gemini to generate images. A lot of the time now I need to generate an image — like one per period for the past three periods. The images are all generated by Gemini and put into the PPT.
Jason: Making the best use of everything.
Wu Wei: OK, time is up — let’s wrap up here. Everyone can keep the conversation going afterward.