Connor Hughes

controls @ sunday

[Connor Hughes @conhugh]

@alpercanbe robotics needs more moose

[Connor Hughes @conhugh]

@kevinmkarol @sundayrobotics 🦾🚀

[Connor Hughes @conhugh]

@kevinmkarol doing it right 👌

[Connor Hughes @conhugh]

@clayhaight chews stuck straight onto the bars is diabolical

[Connor Hughes @conhugh]

@alpercanbe

[Connor Hughes @conhugh]

@ChunhuiLiu96 robots tend to teach that second one real fast! @ChunhuiLiu96 we're lucky you decided to take the leap and come push AI frontiers on real robots 🦾

[Connor Hughes @conhugh]

it's a rare privilege to be bitter-lesson-pilled in robotics

[Chunhui Liu @ChunhuiLiu96]

Four differences I've noticed since moving from LLMs to robotics:

1. The data boundary changed. LLM scaling is obvious because the boundary is clear: the entire digital world. The physical world has no obvious boundary: you can't just cover everything. Data density and diversity need a rethink.

2. Everything must work at once. Imagine evaling an LLM while a GPU silently returns garbage tensors. A robot policy only works when control, hardware, model, and design, all work at the same time.

3. Eval in physical world is so different.

4. Data assumptions break. Even post-bitter-lesson, LLM recipes inject tons of inductive bias (imagine no dedup or domain balancing). In robotics, some of those biases are guaranteed to break.

[Connor Hughes @conhugh]

@johnson_zhsh @sundayrobotics real g's move in silence like champagne

[em m0shouris @emm0sh]

connecting claude to solidworks is like strapping a raptor engine to a 1995 toyota corolla

[Junyao Shi @JunyaoShi]

Personal update: I've joined @sundayrobotics.

Two questions ran through my whole PhD: how to learn from scalable human data, and how to build general-purpose robots.

Trying to answer them convinced me of one thing: general-purpose robots will never come from better models alone. It takes tight iteration across data, hardware, model, control, and evaluation. Every loop you can shorten matters.

My first dinner with @tonyzzhao and @chichengcc turned into a four-hour conversation. I walked away realizing how much we saw eye to eye: scale the data, think full-stack, start from the problem you want to solve instead of the idea you want to win.

So getting to work at Sunday is a dream come true, a place to solve generalization with the full breadth of human data and system-level thinking, and keep chasing the questions I care most about.

After my first month in, two things stand out: Sunday’s full-stack team iterates unbelievably fast, and the energy when everyone is aligned on the same vision is electric. This speed and energy is exactly why what used to feel impossible now feels close.

Home robots, the frontier physical AI in the hands of ordinary people, were long seen as a distant dream . At Sunday, I watch this dream take shape every day. I'm convinced there's real research-market fit here: foundation models and home robots point toward the same north star, generalization, not specialization, because every home is different.

Excited for the zero-to-one moment ahead.

[John Carmack @ID_AA_Carmack]

If you are asking “Why push back against anti-datacenter efforts?” I consider it a tragedy that anti-nuclear efforts largely strangled nuclear power in the US based on vibes, and I don’t want to see that happen to AI. Public opinion matters, and it shouldn’t be ceded unchallenged.

If you are asking “Why should I support AI efforts at all?” I believe we are in the midst of a transition more vibrant than the industrial revolution. Opinions formed a couple of years ago about the uselessness of AI are no longer valid. Millions of people and organizations are getting great returns from using it, and the demand for data centers is the market responding to the value signal. That is how progress is made!

[Connor Hughes @conhugh]

Speedups across the board! maybe a disproportionate boost when integrating new sw components into your stack, adding new logging or visualization, or chasing needle-in-a-haystack bugs. Hard to think of anything totally unaffected.

What hasn't changed is that the only way to define a control policy that actually works across the fractal complexity of real-world environments, objects, tasks, etc, is to use gradient descent.

[Connor Hughes @conhugh]

> "we now seem much closer to a world where models will be able to use off-the-shelf physical tools with relative ease... more research is needed to understand models’ ability to make these physical tools more bespoke, whether by writing control policies tailored to particular tasks or by designing robotic systems"

When this works, it will be bc frontier models can make api calls to other models trained on real-world data at scale. It won't be by writing control policies tailored to particular tasks.

Frontier models are obviously superhuman at code-gen, but code-as-policies has always been doomed. Not bc humans are 10x or 100x or 1000x too slow at writing good code, but bc software 1.0 is simply not the right action space for robotics. This is a key distinction b/w physical and virtual agents.

[Anthropic @AnthropicAI]

New Frontier Red Team blog: Phase 2 of Project Fetch, where we test how well Claude can program a robodog.

Opus 4.7, on its own, was ~20x faster than last year's best human team aided by Opus 4.1. (The robodog, alas, still failed to fetch a beach ball.)
https://t.co/CgbBtRf85e

[Connor Hughes @conhugh]

@ryancjulian @JunyaoShi @huaijiangzhu then sip signal from tiny straw, I'm afraid

[Connor Hughes @conhugh]

fable is amazing but silently nerfing ML research is one big wrinkle

[Connor Hughes @conhugh]

@alpercanbe why this shape?

[Sunday @sundayrobotics]

1,709 → 72,000 sq. ft. in two years

Welcome to Sunday’s new HQ.

[em m0shouris @emm0sh]

to illustrate why the world is held back by engineering software (CAD, etc.):

what would the world look like today if python were paywalled like matlab is, and there were no alternatives?

apply that to everything we do and see the issue

[Sunday @sundayrobotics]

We’re hiring in-house Memory Developers to shape how Memo perceives and interacts with the world.

The Role:
📍 On-site & flexible (3–5 days/week).
👟 Record high-quality task demos (e.g. folding t-shirts, arranging shoes, etc).
🧠 Beta-test new hardware and software.
📈 High growth potential into management.

Apply below if you’re detail-oriented, curious about AI, and interested in contributing to the future of autonomous home robotics.