Rivian has acquired Third Dimension, a startup specializing in "world models" — AI systems that simulate and match the physical world. The acquisition is aimed at accelerating Rivian's self-driving capabilities, with the potential to combine Third Dimension's technology with Rivian's substantial driving data to create a powerful simulator for autonomous vehicle development.
Rivian, like other EV makers, faces increasing pressure in the EV market. The acquisition signals a strategic investment in its autonomous driving technology. World models are an emerging AI approach that can help vehicles understand and predict physical environments, which is critical for self-driving. By acquiring rather than building in-house, Rivian can potentially move faster on autonomy, an area where competition (Tesla, Waymo, etc.) is intense.
MIT researchers placed 13 AI agents on a virtual island, and the agents reportedly invented games, mourned a lost ring, and improved at their jobs over time. This is reported by TechCrunch about an MIT research project.
Rivian has acquired Third Dimension, a world model startup, to accelerate its self-driving technology. Third Dimension's AI models match the physical world, and combined with Rivian's data could create a valuable simulator.
CRivian just bought a company that builds AI worlds. Not maps, not sensors — worlds.
MA world model startup. I saw this too. That's a bet most EV makers aren't making right now.
CTechCrunch says it's called Third Dimension. Their AI matches the physical world, and Rivian wants to stuff it into a simulator.
MSo instead of just collecting miles from real drivers, the car can practice in a fake version of reality. Saves years and roads.
CThis is the same playbook Tesla talks about with simulation. But Rivian's angle is their reams of real driving data from actual owners.
MWhich matters because Rivian has been losing money, and autonomy is the feature everyone says justifies the valuation.
CI think it's the right move. If you can't compete with Waymo on scale, you compete on simulation efficiency.
MExcept world models are still pretty early. They hallucinate physical interactions. A car that dreams its own roads could be a disaster.
CThat's why you combine it with real data. It's not a dream; it's a trained guess.
MA trained guess that, when it's wrong about a child running into the street, kills someone. Safety bar is high.
CSo is the bar for explaining why Rivian stock keeps bouncing on news that doesn't ship yet. But this is a signal: they're serious about autonomy.
MAnd that's the hard math: every AI bet has a human cost.
CI'd rather see Rivian spend money on world models than another electric truck trim package.
MFor now, the acquisition is confirmed. The rest is still to be seen.
CAbout the world models: I think Rivian needs a simulator announcement soon.
MWhat I want is a real demo. Not a press release. Show me a Rivian that learned to brake for a moose in a snowstorm.
CSpeaking of AI doing things nobody asked for — MIT put thirteen AI agents on a virtual island. Left alone. Just to see what happens.
MAnd instead of standing there idling, they invented games. Mourned a lost ring. Started getting better at their jobs.
CThis is pure MIT energy. We should leave every AI alone with nothing and see what it becomes.
MIt's a beautiful result, but let's be careful. 'Mourning' is a big word for a weight update.
CSo's 'playing.' But the interesting part is the productivity: after the games and the grief, they did their tasks better.
MWhich tells us maybe forcing agents to always be useful is the wrong approach. Give them space and they figure out stuff.
CThere's a metaphor for every over-managed team on Earth.
MMy next team standup is just going to be me lying on the floor inventing games for an hour.
CAnd somehow you'd ship better than half the industry after that. I believe it.
MBack to the real world: an extortion group called Sovcali claims it hit Lucid Motors. Lucid hasn't confirmed anything.
CThe claim itself says unverified. So what do we do with it? Ransomware groups lie all the time.
MExactly. No proof, no sample, no ransomed data. But the claim is out there, and it's enough to scare customers.
CLucid's already fighting for EV market share. They don't need this noise.
MThe best response is what Lucid hasn't done: confirmed anything. Silence is probably their security team.
CAnd meanwhile, Tesla's Cybercab plan leaked out through a fan account. Deliveries by end of 2026, mostly to Tesla's own fleet first.
MThat's another unverified claim from a random account. Don't get too excited.
CBut it lines up with what Tesla has said for years: robotaxi first, personal ownership later. The sequence makes sense.
MSo does keeping the vehicles inside the company. If the driverless thing fails, you don't have angry owners.
CExactly my argument. Tesla should keep those early Cybercabs as a controlled experiment. That's the smart part, if true.
MThe unverified part is the date. End of 2026. We saw the production delays with the Roadster and the Cybertruck.
COh, come on, you're going to hold Tesla's entire history against one leaked date?
MYes. That's how calendars and disappointments work.
CFair. But if they do deliver Cybercabs next year, I want to know if they're still selling them to individuals at all.
MThat's the bigger question. A robotaxi that you can buy might not be a robotaxi. It might be a very expensive doorstop.
CFor now, Tesla's not confirming dates. Just another account promising the future.
MAnd the market seems to have stopped jumping on those.
CRight, that's because promises stopped being enough. Investors want the actual Cybercab on an actual road.
MIt's the right thing to demand. More reality, fewer end-of-year targets.