Busted: Elon Musk admits new Optimus video isn't what it seems
Elon Musk has Xeeted out a video of Tesla's Optimus humanoid robot folding a t-shirt – which would be great, but then followed up with a second Xeet clarifying that the video is far less impressive than it looks – after being called out by observers.
Dozens of companies are now pushing to make 2024 the year when humanoid robots take their first wobbly steps into warehouses and onto factory floors, and start doing actual useful work – real tasks that replace human labor.
But in order to do that, they'll have to learn to learn; bipedal walking and yoga balancing are one thing, but being able to take instructions and go and do something useful is another altogether. This is why Figure CEO Brett Adcock claimed a "ChatGPT moment" for humanoid robotics a week and a half ago, when he showed the company's 01 robot autonomously operating a coffee machine, having learned solely by watching a bunch of video of humans doing the same job.
Now, Elon Musk has responded by showing Tesla's Optimus robot doing something a lot trickier: pulling a t-shirt out of a basket and neatly folding it, heralding a future in which these humanoids are staple servants in well-heeled homes, taking care of routine chores and making everyone's lives wonderful until the revolution comes and they start folding the wrong socks together out of spite.
Important note: Optimus cannot yet do this autonomously, but certainly will be able to do this fully autonomously and in an arbitrary environment (won’t require a fixed table with box that has only one shirt)— Elon Musk (@elonmusk) January 15, 2024
The problem is, Optimus isn't folding that shirt by itself. There's a human driving it via telepresence, as Musk noted in the follow-up Xeet above, some 23 minutes later. Or putting it another way, some 20 minutes after "TeslaLisa" pointed out the telepresence control glove poking into frame.
Autonomous vs telepresence was a distinction Adcock was particularly keen to point out – and with good reason.
Which makes it a pretty confusing release from Tesla from where we stand; of course Optimus can do anything from the Macarena to possibly even tying shoelaces with a human operator driving it – the hardware looks capable of all kinds of fine movement controls – as do many of the other humanoids in the race.
But that's not the hard part – or indeed the useful part, or indeed the new part. The big question is "what can Optimus do by itself?"
Over the coming months and years, we're bracing for an avalanche of little videos like this as humanoids across the board start to pick up basic abilities, then more complex tasks, and eventually entire self-directed jobs in dynamic, potentially even chaotic environments.
We are very much at the pre-dawn stage of general purpose humanoids, but as Toyota's stunning "large behavior model" progress shows, embodied AIs are quickly leveling up their ability to watch and learn. Given a bunch of examples of humans doing a given task, they're capable of figuring out what success and failure look like, then running millions of simulations to figure out how to use their own "bodies" to get the same job done, correcting for errors along the way.
That's the exciting bit, and that's what we need to see from Tesla – not what the hardware can do with a human driver at the wheel. And certainly not videos that kinda, sorta, maybe imply autonomy, only to clarify that it's human-operated 20 minutes after being called out by eagle-eyed observers on the Xitter.
To be clear, Tesla has serious chops in AI and autonomy, and we don't expect it'll be long before Optimus starts buttering its own toast and mastering the electric drill and the egg beater. But this one's just not a good look.