Pinned Tweet
Two months of ablating π0.5 finetunes on a real manufacturing task. The policy now succeeds 98% of the time.
We're publishing all results, all the data, and every run, including the ones that went nowhere.
What we learned:
1/ 5x more data took us from 63% to 76%. Pure scaling e found.
2/ Diversity got more out of the same hours. 4h spread across five scenes beat 4h in the eval scene by 30pp.
3/ Quality did even better. One clean, slower hour on top of the 21h: 76% → 90%. That single hour added more than the 17 hours before it.
4/ So we tried quality on its own: 1h of clean data plus 240 rollouts with human interventions. 1.7h in total, 28% → 88%. That's 12 points above the 21h model with a twelfth of the data.
5/ Inference settings add more on top, with no retraining. They took the 21h model from 76% to 93%, and the 21h+1h model from 90% to 98%.
Just as useful: what didn't matter. Learning rate, batch size, relative vs. absolute joints, image augmentation. All less important than we expected.
Full write-up linked below.
We're publishing all results, all the data, and every run, including the ones that went nowhere.
What we learned:
1/ 5x more data took us from 63% to 76%. Pure scaling e found.
2/ Diversity got more out of the same hours. 4h spread across five scenes beat 4h in the eval scene by 30pp.
3/ Quality did even better. One clean, slower hour on top of the 21h: 76% → 90%. That single hour added more than the 17 hours before it.
4/ So we tried quality on its own: 1h of clean data plus 240 rollouts with human interventions. 1.7h in total, 28% → 88%. That's 12 points above the 21h model with a twelfth of the data.
5/ Inference settings add more on top, with no retraining. They took the 21h model from 76% to 93%, and the 21h+1h model from 90% to 98%.
Just as useful: what didn't matter. Learning rate, batch size, relative vs. absolute joints, image augmentation. All less important than we expected.
Full write-up linked below.
55
63
708
79,630
Last day in Shenzhen. We're spending it at the Powerhouse office in Shenzhen’s Bao'an district.
Ryan moved here 11 years ago to build his own product. He then started Powerhouse to help Western companies build and scale hardware production in Shenzhen.
The screen behind us was ts they built for a client: a generative art display for hotel lobbies and very rich people. It took me a while to notice it's moving (second video is more obvious).
He’s an expert on getting hardware built in China and he’s hosting a first batch of young founders who want to build hardware here at his office soon. No cost. If this is you you should reach out to him!
Ryan moved here 11 years ago to build his own product. He then started Powerhouse to help Western companies build and scale hardware production in Shenzhen.
The screen behind us was ts they built for a client: a generative art display for hotel lobbies and very rich people. It took me a while to notice it's moving (second video is more obvious).
He’s an expert on getting hardware built in China and he’s hosting a first batch of young founders who want to build hardware here at his office soon. No cost. If this is you you should reach out to him!
Reinforcement learning, AI/ML for robotic arms, building in public vs stealth, when to bootstrap vs raise VC money.
Just some of the topics in our Sunday afternoon powerhouse chats with @DominiqueCAPaul @aurel_arnold @_marcobene_ over baguettes, avocado, cheese, blueberries and MeetLayer generative art display giving good vibes in the background.
Perfect Sunday 🔥
8
7
117
8,050
Dominique Paul retweeted
I’ve moved to china 🇨🇳 , and I’ll be spending quite some time here.
I spent the past few months in San Francisco. Now I’m in Shenzhen.
I’m convinced we’re entering a new founder era.
Not “move to SF to build.”
But “move to Shenzhen to build.”
Here, ideas, engineers, s are all within reach.
The startup world is shifting. And founders should pay attention.
I spent the past few months in San Francisco. Now I’m in Shenzhen.
I’m convinced we’re entering a new founder era.
Not “move to SF to build.”
But “move to Shenzhen to build.”
Here, ideas, engineers, s are all within reach.
The startup world is shifting. And founders should pay attention.
256
54
1,304
117,010
This feels so futuristic. Shenzhen does a 10-minute lightshow several times on weekends where they coordinate the 1.2m LEDs on 43 towers.
I have no idea how they‘re even able to connect and centralize the controls for so many buildings.
I have no idea how they‘re even able to connect and centralize the controls for so many buildings.
24
8
259
13,059
I think I found some of them today
@DominiqueCAPaul @Fateen_Anam theres a reason why almost all global stolen phones make their way to HQB
2
1
10
2,045
Dominique Paul retweeted
The “Move to Shenzhen” Era: the theme of my talk tonight. Full video coming soon.
Met lots of my new X friends IRL too: @DominiqueCAPaul @imittalishika and many more. Shenzhen events are booming, so…
MOVE TO SHENZHEN
IT WILL CHANGE YOUR LIFE
(Seriously)
Met lots of my new X friends IRL too: @DominiqueCAPaul @imittalishika and many more. Shenzhen events are booming, so…
MOVE TO SHENZHEN
IT WILL CHANGE YOUR LIFE
(Seriously)
At 23yo, I moved into a Chinese factory and became 杜无敌. Eleven years later, someone’s giving me a microphone to explain myself.
This post is now happening IRL.
Factory stories, expensive mistakes and a few things that actually worked. Hosted by Josh Woodard at Tech in the , Shuiwei, Shenzhen
🗓 Saturday 24 October, 6pm
Free entry. Come learn from my mistakes. Much cheaper than making your own.
12
6
81
4,246
Ryan moved to Shenzhen 11 years ago to build hardware products. Now he started a space to help startups get from idea to factory floor, with an engineering and manufacturing team on the ground in Shenzhen.
The other surprise is how small the SZ X bubble is. Same faces at every ee so many people IRL after reading about their experiences in the feed.
The other surprise is how small the SZ X bubble is. Same faces at every ee so many people IRL after reading about their experiences in the feed.
At 23yo, I moved into a Chinese factory to build hardware for my first startup.
That’s where I got my Chinese name: 杜无敌 (Duwudi). “Du the Invincible.”
Eleven years later, I’m still in Shenzhen. Apparently I haven’t suffered enough.
Along the way: $40M raised, 500K+ custom Guinness World Record for the world’s first 3D-printed laptop.
Now I run Powerhouse out of an old factory, helping other founders get from idea to mass production at Shenzhen speed. Ideally with fewer expensive lessons than I had.
We’re also launching a hardware accelerator, and I’m writing my first angel cheque. The plan: build and back the next generation of hardware companies.
I share what building hardware here actually looks like: prototypes, production problems, factory life and robots behaving badly.
Who should I be following in hardware, physical AI and robotics? If that’s you, say hello 👋
13
1
110
8,232
Dominique Paul retweeted
We're excited to announce ZurichCV#15 is next Thursday at the @ETH_AI_Center!
Said Gürbüz (ETH Zurich, IBM Research) will present DeskForge: Controllable Desktop Environments for Training Computer-Use Agents.
Lorenzo Steno @lsteno (Seldon Technologies @seldon_tech) will talk about building RL environments for computer use.
RSVP: zurich.tech/events/zurichc…
(PS: this is our new account for zurich.tech. please help out by sharing!)
Said Gürbüz (ETH Zurich, IBM Research) will present DeskForge: Controllable Desktop Environments for Training Computer-Use Agents.
Lorenzo Steno @lsteno (Seldon Technologies @seldon_tech) will talk about building RL environments for computer use.
RSVP: zurich.tech/events/zurichc…
(PS: this is our new account for zurich.tech. please help out by sharing!)
2
6
34
13,234
TroubleMaker must be the vibiest makerspace I’ve been to. It‘s in Shenzhen‘s tech and electronics district, Huaqiangbei and was started in 2016 by Henk Werner after his hardware startup in Holland went bankrupt.
He thought most Shenzhen makerspaces existed to collect government with no grants, sponsors or commissions on what members manufacture.
It’s basically a flexible membership space (~$600/month) where foreign builders can come to iterate on their hardware and connect with others doing the same. All kinds of tools and machines are already there and literally any electronics part you need is in a 2 km radius.
He thought most Shenzhen makerspaces existed to collect government with no grants, sponsors or commissions on what members manufacture.
It’s basically a flexible membership space (~$600/month) where foreign builders can come to iterate on their hardware and connect with others doing the same. All kinds of tools and machines are already there and literally any electronics part you need is in a 2 km radius.
11
10
190
8,413
Met @Fateen_Anam in SZ today! I remember seeing this video last year ago and thinking how this one of the coolest AI consumer products I’ve ever seen.
Kids having AI friends isn't just coming—it's already here
0
2
25
2,859
10 PM in Huaqiangbei, SZ.
We’re on our way getting dinner with @Fateen_Anam and Aurel wonders if anyone could swap the battery in his old iPhone.
First stall we ask does it on the spot in 10 minutes. 20€ for a third-party battery, 40€ for an original Apple one.
We’re on our way getting dinner with @Fateen_Anam and Aurel wonders if anyone could swap the battery in his old iPhone.
First stall we ask does it on the spot in 10 minutes. 20€ for a third-party battery, 40€ for an original Apple one.
49
17
500
84,562
Just visited Tianjin Robotics in Shenzhen as we’re moving into our first deployments and need more robust arms. Tianjin builds 2,000+ arms a month, and every arm sells for about $15k.
AGIBot has bought roughly 20,000 of them, which surprised me. I thought they built their own. . NVIDIA asked for a custom version with a longer upper arm, which apparently leads to fewer collisions.
Two most interesting notes for me today:
1/ Camera cable reliability is an industry-wide problem right now. 2 of our 8 arms had it. I liked routing cables inside the arm because external cables look terrible, and assumed we'd just bought bad ones. If everyone has this problem, the cables belong outside, where you can swap one in a few minutes.
2/ They say customers keep asking for higher precision. A learned policy corrects as it goes, so repeatability should matter less for us. Either I'm missing something or most of their buyers aren't doing robot learning.
AGIBot has bought roughly 20,000 of them, which surprised me. I thought they built their own. . NVIDIA asked for a custom version with a longer upper arm, which apparently leads to fewer collisions.
Two most interesting notes for me today:
1/ Camera cable reliability is an industry-wide problem right now. 2 of our 8 arms had it. I liked routing cables inside the arm because external cables look terrible, and assumed we'd just bought bad ones. If everyone has this problem, the cables belong outside, where you can swap one in a few minutes.
2/ They say customers keep asking for higher precision. A learned policy corrects as it goes, so repeatability should matter less for us. Either I'm missing something or most of their buyers aren't doing robot learning.
22
16
245
20,187
Scooters are next level big in China.
This is in front of the gates of a manufacturing company in Songgang.
This is in front of the gates of a manufacturing company in Songgang.
6
0
36
5,251
Dominique Paul retweeted
Maaaan, I was so retarded to fall for western propaganda 😭
What Valentin said, got blown away. So so much to learn. Not to mention if u in hw, the rapid prototyping u can do here just because almost any impossible hw component you need is 30 mins away by car. So u can do like 1y of prototyping u would do while u sit in Europe or usa would be done here in 3 months.
Highly recommend you to visit it and happy to connect with people that helped us to organize the tour.
Special thx to our @tongnk for getting us here, and opening our eyes ❤️
What Valentin said, got blown away. So so much to learn. Not to mention if u in hw, the rapid prototyping u can do here just because almost any impossible hw component you need is 30 mins away by car. So u can do like 1y of prototyping u would do while u sit in Europe or usa would be done here in 3 months.
Highly recommend you to visit it and happy to connect with people that helped us to organize the tour.
Special thx to our @tongnk for getting us here, and opening our eyes ❤️
I spent the last week and a half in China in the Greater Bay Area and I can confirm. Every entrepreneur from the west should come and visit for at least a week. Go on the expos and visit factories and incubators. The level of innovation and speed of development are incredible.
7
5
105
16,286
Siddhartha tries retardmaxxing. He becomes a merchant, drinks, gambles, and twenty years later can't stand himself and as as a consequence wants to drown himself in a river.
Yep, action beats inaction nine times out of ten. But that doesn't rule out reflecting.
Knowing what you actually want is an edge over everyone running on mimetic desire.
Taking the time to figure out what I really care about working on was the best decision I've made - even though retardmaxxing would have been a more comfortable life decision with less struggle in the short term.
The biggest risk of retardmaxxing is spending a decade, or your whole life, on something you don't care about, and only noticing once it's too late to change.
Yep, action beats inaction nine times out of ten. But that doesn't rule out reflecting.
Knowing what you actually want is an edge over everyone running on mimetic desire.
Taking the time to figure out what I really care about working on was the best decision I've made - even though retardmaxxing would have been a more comfortable life decision with less struggle in the short term.
The biggest risk of retardmaxxing is spending a decade, or your whole life, on something you don't care about, and only noticing once it's too late to change.
Is retardmaxxing the right philosophy, or not?
Lots of Europeans have missed out on @pmarca praising himself for having zero introspection, and the great philosophy of retardmaxxing.
I found it very irritating, and some statements are certainly historically incorrect. Like "great men of history did not have introspection"... vs. Socrates "The unexamined life is not worth living".
However, after recently starting to read Siddhartha by Hermann Hesse before bed, I woke up the next morning and shockingly realized that the book reminded me of ... retardmaxxing. Grossly simplified, it's about more doing and lived experience rather than overthinking.
I briefly felt ashamed for not launching my new company 3 months earlier, almost became self-aware, then remembered @JensenHuang telling us Germans to "just get to work", and immediately started blasting through customer messages... zero introspection!
As a business philosophy, this might actually work.
Honestly, when I would reflect on all the things that I could worry about, from European health data compliance to an AI agent going rogue to my mum thinking I lost my mind chasing dreams... I'd get nothing done.
While I think a great life with happy relationships requires a lot of introspection, building a company might not.
Pick a problem you care about, start selling a solution, start building the solution.
The business world is getting re-assembled right now. If you are standing on the sideline wondering what it'll look look in 10 years...
Just get to work!
2
1
22
2,145
Yesterday we visited Shenzhen InnoX Academy, a hardware incubator founded by Li Zexiang, often called the godfather of robotics in China. His student Frank Wang started DJI out of his HKUST lab, and Li backed it early.
Most hardware incubators give you access to 3D printers and o has stress testing rigs for corrosion, drop tests and sustained use, so you find out how your product fails before shipping.
But what I found most interesting is that they help with the first step of manufacturing. Going from a prototype to your first production batch is hard for a lot of hardware startups and where it's easy to lose speed. InnoX has manufacturing people on site who build that first larger batch for you AND help you write the standard operating procedures to go to efficient larger scale production.
They also have a large network of manufacturers, and some of them invest in kind by putting up the capital and the people to make your product. For a hardware startup, that can mean your seed round isn’t only cash but someone providing you with 35 of their workers and factory space to produce your first 100k units.
Most hardware incubators give you access to 3D printers and o has stress testing rigs for corrosion, drop tests and sustained use, so you find out how your product fails before shipping.
But what I found most interesting is that they help with the first step of manufacturing. Going from a prototype to your first production batch is hard for a lot of hardware startups and where it's easy to lose speed. InnoX has manufacturing people on site who build that first larger batch for you AND help you write the standard operating procedures to go to efficient larger scale production.
They also have a large network of manufacturers, and some of them invest in kind by putting up the capital and the people to make your product. For a hardware startup, that can mean your seed round isn’t only cash but someone providing you with 35 of their workers and factory space to produce your first 100k units.
24
26
323
18,139
Just went for a walk in a Hypershell exoskeleton.
For the first stretch I wasn't sure how necessary/helpful it is. Then I switched it off and thought it was actually quite nice, especially on stairs.
Feels like something you'd buy for your parents, so they can come along on a hike.
For the first stretch I wasn't sure how necessary/helpful it is. Then I switched it off and thought it was actually quite nice, especially on stairs.
Feels like something you'd buy for your parents, so they can come along on a hike.
9
5
56
5,970
From the outside it looks like an over-futuristic office, but it's a furniture factory. Today we visited Coomo in Houjie, a town in Dongguan which is China's biggest furniture cluster.
It’s an example of Chinese manufacturers not just trying to win on price but pushing more t. A single armchair costs around €1,500.
6,000+ employees, 14 factories, about 1 million m² of factory floor.
The site used to be a shoe factory. COOMO bought the land and built this crazy complex on it. Production happens inside, a few floors are set aside as showrooms, and the roof is terraces and gardens.
Some notes i took during our visit:
> Work runs 8–12, then a 2-hour lunch, then 2–6. Eight hours a day.
> No fixed weekends. Instead everyone gets 4 days off a month to take whenever they like, and can work extra shifts if they want. Common in China, people want to earn more.
> they regularly host rooftop parties with jazz bands and barbecues for employees. There was one the week before we came.
>They farm vegetables on their rooftop: 200 plant varieties, 6 full-time gardeners, and the vegetables go into the cafeteria.
Also a big employee benefit: They help employees get a local hukou.
Hukou is China's household registration system. Yours is tied to where your family is registered, usually your hometown, and it decides where you get access to public schools, healthcare and social benefits.
A migrant worker in Dongguan can get a residence permit, but without a local hukou their kids face restrictions in local schools and exams. Getting one is slow and conditional. Coomo uses its standing as a large employer to sponsor workers' applications, and in practice the worker's hukou is often held in the company's collective registry.
Leaving the job means moving it out, so it's both a real perk and a strong reason to stay.
We met two Austrian real estate professionals here. Their take is that some products are more fitting for the Middle East and wouldn’t work for Europe, but the ones that do are extremely high quality and still significantly less expensive compared to their western equivalents.
It’s an example of Chinese manufacturers not just trying to win on price but pushing more t. A single armchair costs around €1,500.
6,000+ employees, 14 factories, about 1 million m² of factory floor.
The site used to be a shoe factory. COOMO bought the land and built this crazy complex on it. Production happens inside, a few floors are set aside as showrooms, and the roof is terraces and gardens.
Some notes i took during our visit:
> Work runs 8–12, then a 2-hour lunch, then 2–6. Eight hours a day.
> No fixed weekends. Instead everyone gets 4 days off a month to take whenever they like, and can work extra shifts if they want. Common in China, people want to earn more.
> they regularly host rooftop parties with jazz bands and barbecues for employees. There was one the week before we came.
>They farm vegetables on their rooftop: 200 plant varieties, 6 full-time gardeners, and the vegetables go into the cafeteria.
Also a big employee benefit: They help employees get a local hukou.
Hukou is China's household registration system. Yours is tied to where your family is registered, usually your hometown, and it decides where you get access to public schools, healthcare and social benefits.
A migrant worker in Dongguan can get a residence permit, but without a local hukou their kids face restrictions in local schools and exams. Getting one is slow and conditional. Coomo uses its standing as a large employer to sponsor workers' applications, and in practice the worker's hukou is often held in the company's collective registry.
Leaving the job means moving it out, so it's both a real perk and a strong reason to stay.
We met two Austrian real estate professionals here. Their take is that some products are more fitting for the Middle East and wouldn’t work for Europe, but the ones that do are extremely high quality and still significantly less expensive compared to their western equivalents.
8
9
159
10,608