controls @ sunday
San Francisco
Joined November 2025
- Tweets212
- Following542
- Followers646
- Likes4.1K
in 2017, a robotics student hears of a sermon by the great Karpathy-sensei
he learns, "gradient descent is a better programmer than you", and is filled with awe
the student repeats this mantra and returns to the lab, chops wood and carries water
...
one night 9 years later, the student sees a sudden apparition
distraught, he races up the ten-thousand steps to the temple
breathless, he asks, "is gradient descent a better programmer than Astra?"
Karpathy-sensei sits in silence for a moment
then sends the student back down the mountain
Connor Hughes retweeted
Sunday Robotics' Tony Zhao runs a fully autonomous robot demo live and unscripted, and it corrects itself every time you mess with it:
@brexton: "I want to confirm for the audience right now that this is fully autonomous. There's no teleoperation, no one behind the curtain, and this is real time. And we did not prepare this demo before. Why do you think seeing real-time demos is so rare for robotics? What the audience is seeing right now is not that common."
@tonyzzhao: "In order for a live demo to work, it's really stressing all parts of the system. From the bottom is the reliability of the hardware, then the software system, then the model. Because we're a full-stack company, we're able to do excellent work across all of them. We have these robots running pretty much every day, and it just keeps getting better as they practice."
@brexton: "So kids can bump into Memo, people can drop clothes, and it will correct itself just on the fly."
@tonyzzhao: "Act Two is the first robotic policy that can handle the diversity of homes in a way that is this reliable. You can see as I'm twisting it, if I twist it in again, it will keep doing it. So Memo has infinite patience."
@sundayrobotics
Connor Hughes retweeted
me, begging, crying, on my knees: "Please just use plain english, I don't understand what you're saying."
Claude: "The right fix, and the book's lesson applies: the tutorial rotted as a front door because a front door full of claims always rots. So the new root holds only what's timeless — the thesis and three doors — and every claim lives behind it in the thing that goes red when stale."
It’s a rare privilege to be bitter-lesson-pilled in robotics.
You often hear voices in robotics saying “scaling up X-type data won’t solve Y”, and without fail, the folks saying this have not actually tried scaling up the data.
I don’t blame them. It takes an incredible amount of work to get high-quality data at scale. Once you have this, though, it’s amazing how it helps cut through the noise and reveal the research problems that matter.
The research Sunday does would not be possible at all without the extraordinary efforts of our 1,000 Memory Developers and our elite Ops team. I’m grateful that I get to help build on their work.
Connor Hughes retweeted
1/N Two years ago, we recruited our first Memory Developer off Craigslist and onboarded them in a public library.
Today, more than 1,000 Memory Developers have helped us build the data engine behind ACT-2.
Here's how we got from that library to here 🧵
Connor Hughes retweeted
We just hit a weird milestone: our model became more reliable than your average home WiFi.
Just like everybody else, we thought cloud inference was the obvious choice. Yet 2 days into the ACT-2 eval, our mind completely changed.
If our hero @ArpitKalla didn’t cook, this video wouldn’t exist 🧵
This video is a great snapshot of our progress, and the blog post offers an exciting glimpse into the scaling laws we’re discovering. For me though, some small, unintentional, unscripted moments have done the most to make our progress feel real in recent months. I'll share anecdotes from a couple of those moments in this 🧵
(6/7) Whoever started it must have just left it running and gone home, I guess? The robot had been running unsupervised right in front of us for half an hour, and nobody noticed. In fact, a few minutes prior, it had finished a whole pile of laundry, and @ChunhuiLiu96 had even given it some more clothes to work through, assuming one of us was running a long test. This robot had accumulated multiple stacks of folded clothes. Memo was ready for more.
(7/7) It’s never been possible before to make this particular mistake. You could not simply set a robot to the task of doing a routine household chore and then forget about it and leave.
The idea is that one day soon, our users will do just that, all the time. In the 11 months I’ve been at @sundayrobotics, it feels like that day has gotten about 3 years closer.
Connor Hughes retweeted
Three years ago, when I witnessed the breakthrough of LLMs, I became convinced that a similar inflection point would eventually arrive in robotics.
Over the past several years of robotics research, I’ve learned that a single overfit demo will not reveal robotics’ “GPT moment”—and that even many impressive demos do not add up to physical AGI. What we need is a robot brain that can perform with human-level quality and robustness across the long tail of real-world situations.
That was why I subscribed to ChatGPT three years ago: it was the first time AI felt broadly useful, not merely impressive.
Now comes ACT-2—a meaningful step toward bringing the same transition to robotics: from demos to generalization, from isolated capabilities to scalable intelligence, and from impressive videos to systems people can actually rely on.
This fall, ACT-2’s first Solve enters homes through our Beta Program, the final step towards fully autonomous home robot deployment.
Full technical report: sunday.ai/blog/act-2-preview
Connor Hughes retweeted
I go to work every day excited about which skill we'll "Solve” next. Many of our future skills “Solve”d will involve the sense of touch, allowing Memo to perform even forceful tasks gently, and safely. It is a blessing to be a part of this exciting journey.
More to come!
Connor Hughes retweeted
@tonyzzhao we forgot to connect Memo to the wifi!
Oh wait we don’t need it.
We moved our robot policy to run completely on-device. Why? Because “works great until the WiFi hiccups” is not an acceptable failure mode for a robot operating in your home. Local inference: predictable latency, no surprises.
