14 September 2026
Heard In AI

A robot said to outrun Usain Bolt splits a panel over what bodies are for

On Moonshots, Peter Diamandis described a new Unitree humanoid as hitting 12.66 metres per second after three months of development. Salim Ismail said the lesson is to stop copying humans; Alex argued general-purpose robots will swallow the specialists; Emad Mostaque predicted extreme machines will be kept off the street, and Alex proposed consumer, industrial and military classes.

A briefing reports one development when it happens. We correct or clarify it later; a new development gets a new briefing. How our formats work

Peter Diamandis introduced the robotics segment of his Moonshots panel with a clip and a set of numbers: Unitree's newest humanoid, he said, was only three months in development and had broken "every human standing jump and speed record" — a two-metre standing jump and a top speed of 12.66 metres per second, against the 12.4 metres per second Usain Bolt reached inside his 9.58-second 100-metre world record. He queued up a second video of the sprint, he said, because it "ends in a nice little scenario here." As the machine ran, a voice on the panel supplied the punchline: "Oh my God, it needs brakes."

These are the figures and the footage as presented on the show. What followed was not a discussion of the record but an argument about what a robot body is for, and who should be allowed to walk one down a street.

Stop making them human

Salim Ismail went first, and went blunt. "I think we should stop trying to make robots human," he said. "Just make them economically useful." His reasoning is that the human body is not an engineering optimum but the result of four billion years of pressure "to survive and procreate" — a shape tuned for staying alive, not for doing a job. If you want a mining robot, he argued, build a mining robot: give it wheels, give it multiple arms, give it whatever the task requires. (He also thanked listeners for sending him pictures of six-armed robots.)

The part of the demonstration Salim found genuinely significant was the calendar, not the sprint. Three months from development to a machine like this is a compression of the hardware iteration loop, he said, driven by several curves multiplying at once — AI, simulation, batteries, actuators. Hardware, on his reading, is starting to move at something closer to the pace of software.

The general-purpose counter-argument

Alex had been reading up on how Unitree got there, and said that on publicly available information the team shifted the robot's mass budget toward the legs, subtracting mass from parts of the upper body. Someone on the panel had a name for it: the robot was "leg maxing" — optimising hard for the thing being measured, in the way a student crams for one exam.

That, Alex argued, is exactly why the specialist future Salim describes will not hold. A world of robots with powerful legs and weak arms, alongside other robots with powerful arms and no human shape at all, is "not a stable equilibrium" in his view. His analogy came from the 1970s and early 1980s, before general-purpose personal computers arrived: dedicated word-processing machines and other single-function devices, with Wang in Massachusetts among the names in that business. General-purpose computers subsumed them. Alex expects the same pattern in hardware, mirroring what he sees happening with generalist AI models swallowing specialist ones: "They're going to be general purpose, and they're going to be general body plan, and they'll be good at everything would be my bet."

"Just put wheels on it," came the interjection from Salim's side of the argument. Alex's reply was that wheels are precisely not general purpose, and he reached for Doctor Who: the Daleks originally could not climb stairs, and only in the newer series can they. Generally capable, for him, means legs in the short term — and, he added, perhaps nanites in the long term, a nod to the microscopic machines of Neal Stephenson's The Diamond Age.

Dave treated the same question as an investment case rather than a design dispute. Robotics is fertile ground precisely because there are so many form factors and sizes to try, he said, with AI-assisted mechanical design now working well enough that you can "just vibe up parts," and manufacturing supply chains attracting serious investment for the first time in decades.

"These types of robots will be banned from the streets"

Emad Mostaque took the demonstration somewhere else: "I think that these types of robots will be banned from the streets." A machine that is super strong and super fast is, by definition, beyond human capability; put enough of them in public and you get accidents and issues, he argued, "just like cars."

Diamandis pushed back on the premise. There will come a point, he said, when these robots are running an AGI model and can avoid accidents — a system capable enough to handle the situation the way a person would, or better. So is the objection that they are not trustworthy? What would actually keep them off the streets?

Mostaque did not concede the capability point; he narrowed his claim. The extreme robots — the ones with beyond-human capabilities — will be kept off the streets or regulated, he said. "That's kind of my contention here." What that opens up, on his account, is a market for soft robots: he pointed to 1X, whose machines some on the panel are due to receive, as "nice and soft" and unable to twist off someone's head or accidentally punch a hole in them. Most people, in his framing, will end up with the Volkswagen of robots while the Ferraris exist elsewhere.

The manufacturer's own description runs along those lines. 1X describes NEO, its home humanoid, as built around a deformable polymer lattice over its internal hardware, with covered joints that keep fingers out of pinch points and tendon-driven actuators that keep the moving mass low. Those are the company's claims about its product, and they describe the mechanics of contact with people rather than speed or strength.

Classes of machine

Alex, who had spent the previous minutes arguing for one universal robot, said he agreed with Mostaque on this. He compared it to rules that already sort machines by what they can do to their surroundings: different regulations for trucks, cars and motorcycles, limits on which roads can carry what, and laser power classes split at the five-milliwatt line. In the near future, he expects roads zoned for a given power density of robot, and classes of machine — consumer-grade, industrial and military-grade — separated by power or torque density. "Totally buy that," he said.

No such zoning or robot class exists; it is a prediction made on a podcast by two people who disagree about body plans and agree about brakes. The question the sprint leaves behind is not how fast a humanoid can go, but what it is allowed to do on a pavement shared with someone who cannot get out of the way.

Share this article

Go to the original

Sources & further reading

  1. 01

From the conversation

Podcast episodes

Article history

Updates to this article

Tags

Altman says economic inertia slowed AI's impact. His podcast panel disputes the cause

On Moonshots with Peter Diamandis, the panel watched Sam Altman explain that he expected GPT-4 to put software businesses up for grabs far sooner than it did, and that the economy's inertia has made the transition "smoother and slower." Salim Ismail blamed institutions that move at a different speed from the technology, Alex pointed instead at the abstraction layers of the economy and prescribed vertical integration, and Emad Mostaque objected that the models simply were not good enough until recently.

6 min read

Uber's drone deal asks whether owning demand beats owning the robots

Uber is investing in Zipline and putting drone delivery into Uber Eats, with a target of one million deliveries a day by the end of 2029. On Moonshots with Peter Diamandis, the panel treated it as the clearest test yet of Uber's strategy of aggregating customers while other companies build and fly the machines — and of what happens if those suppliers decide to sell direct.

5 min read

OpenAI paused some frontier training, and the panel split over why

OpenAI said on August 18 that it had halted part of its frontier reinforcement-learning work until alignment, security and monitoring standards caught up with the capabilities ahead. On Moonshots with Peter Diamandis, Emad Mostaque called the safety constraint real, Alex called the announcement marketing, and Salim Ismail reported back from a visit to OpenAI's offices. OpenAI later disclosed that the largest paused run restarted on August 28.

9 min read

A 100x claim lands, and the panel hits a harder question: coordinating 10,000 agents

On Moonshots, the panel revisits Elon Musk's January prediction that models were "off by two orders of magnitude" in intelligence per gigabyte, after Tim Sweeney tweeted that it had come true and Musk replied that specialist AIs add another 100x. Dave calls 100x a lower bound and asks what anyone would actually do with 10,000 brilliant agents; Emad Mostaque describes running specialized agent teams, while Alex argues Musk's "specialist models" are really sparsification inside generalist models.

6 min read