OpenAI and Elon Musk have a long and acrimonious history, but they now seem to be bickering over aspects that are not directly related to AI.
Peter Welinder, VP of Product at OpenAI, put out a lengthy post on X laying out why he believes affordable Level 5 self-driving is still 5-10 years away. Musk, whose entire Tesla valuation thesis rests on the opposite claim, replied with a single clown emoji. No quote tweet, arguments, or data — just the emoji, sitting there under a post that had taken the time to define exactly what it was arguing against.
Welinder’s argument starts from a premise that will annoy Tesla bulls before they even get to his bar for Level 5. He says affordable Level 5 requires AGI, and that AGI itself hasn’t arrived yet. His logic runs in steps: first you need multimodal, multiplexed AGI, then real-time inference on top of that, and only then can any of it be squeezed onto an energy-efficient, low-cost chip that could sit inside a car people can actually afford. “Many steps left,” he wrote, in what reads like a deliberately understated way of saying the industry isn’t close.
What made the post spread beyond the usual AI Twitter circles was the second half, where Welinder tried to make his claim falsifiable instead of just rhetorical. He set out a specific bar for what “affordable Level 5” should mean: a car priced at $30,000 or less, or a robotaxi ride costing under $0.65 per occupied mile. No geofence, no remote operators and no dependence on GPS or centimeter-accurate HD maps — just a normal map and onboard perception, the way a human driver operates. The car should be able to go anywhere on Earth a competent human could drive, handle weather, bad road markings, construction, police hand signals, and the informal negotiation that happens at every four-way stop in a city that hasn’t repainted its lines in a decade. And it should complete an entire trip end to end, including recovering when something goes wrong, without ever asking a person to take the wheel.
It’s a description that reads almost like a direct rebuttal to how Tesla’s robotaxi program actually works today. The service in Austin and the Bay Area has leaned on geofenced routes and, for stretches of its rollout, safety monitors either in the driver’s seat or trailing the car — arrangements that Tesla has quietly reintroduced after briefly attempting to run without them. Reports have also pointed to a rising rate of incidents as the fleet has scaled, even as the number of vehicles involved remains a fraction of what Waymo already operates. Whether Musk’s emoji was aimed at Welinder’s AGI premise, his cost bar, or the entire framing of the post is impossible to know, since he didn’t say.
This isn’t the first time Musk has used the same move against a rival’s technical claims. When Nvidia unveiled its Alpamayo self-driving models, Musk didn’t dismiss it outright — he argued instead that Nvidia would discover what Tesla already had, that getting to 99% reliability is easy and the last percent is where years get spent. That was at least a counter-argument. This time, he skipped the argument and went straight to the punchline.
The OpenAI-Musk relationship has been fraying for close to a decade now, ever since Musk left the company he co-founded and watched it turn into the industry’s most valuable AI lab without him. He’s since sued OpenAI more than once, lost a jury trial against it in May on a technicality around timing rather than merits, and traded insults with Sam Altman over everything from the nonprofit-to-for-profit conversion to, more recently, calling him a scammer over an unrelated Apple lawsuit. Altman, for his part, has called Musk a bully who “likes to get in fights,” listing off Bezos, Gates, and Zuckerberg as fellow targets before landing on himself.
Whether Welinder is right about the timeline will take years to settle, not a single social media reply. But the fact that a post about chip efficiency and mapping requirements turned into the latest flashpoint between Tesla and OpenAI says something about how personal this rivalry has become. At this point, Musk doesn’t need a reason connected to AI to take a shot at anyone associated with the company. A post about robotaxis will do just fine.