MA ride-hailing company nobody talks about just jumped into the robotaxi race, and they're not testing in California or Texas.
CWhich company?
MinDrive, the Uber rival that's huge in developing markets. TechCrunch reports they're teaming up with Pony.ai and Nvidia to test robotaxis in Kazakhstan's capital in early 2027.
CKazakhstan. That's... actually pretty smart.
MThink about it. Waymo is fighting San Francisco traffic. Tesla's fighting American regulators. inDrive is going somewhere that might mean lower labor costs and less red tape.
CBut the tech has to actually work. Pony.ai has a decent record in China, but Central Asian roads? Winter in Astana? That's a different beast.
MAnd that's exactly why they're testing there. If your autonomous stack survives a Kazakh winter, it survives anywhere.
CI'll give them credit: they're not doing the boring Silicon Valley demo loop. This is real-world, messy conditions, and if it works, they undercut everyone.
MMeanwhile, the big names are still fighting over whether robotaxis can even turn a profit.
CDifferent world entirely: MIT researchers left thirteen AI agents alone on a virtual island with nothing to do.
MAnd they invented games. And mourned a lost ring.
CTechCrunch has the story. The researchers gave them a simulated world, no tasks, no goals, and just watched what happened.
MSo the AIs got bored and started... playing? That's adorable and a little creepy.
CIt's not adorable. It's emergent behavior. When they had nothing to do, they created their own objectives. That's how you get agents that don't need to be told what to do.
MEmergent behavior? You just described a toddler who found a cardboard box.
CBut a toddler with a cardboard box becomes an engineer. The MIT paper says they got better at their actual jobs after the play session.
MSo unstructured downtime makes AI more productive. Great. Now HR is going to make us all do mandatory fun.
CYou joke, but this is a real insight. Agents that can invent their own tasks are the ones that'll handle messy real-world jobs.
MSpeaking of messy real-world jobs: Rivian just bought a startup called Third Dimension to make their self-driving software less bad at physics.
CLess bad at physics. That's the technical term.
MTechCrunch reports the deal gives Rivian world-model tech, which basically means AI that understands how objects move in the real world. Cars don't teleport. Rain makes things slippery.
CAnd Rivian has mountains of real driving data. Combine that with a model that actually simulates physics, and you get a training loop that isn't just watching dashcam videos.
MBut here's my question: why is Rivian spending money on an acquisition right now?
CBecause they can't afford to be left behind on autonomy. If they don't have a credible self-driving story, they're just a truck company with nice headlights.
MA truck company with nice headlights. Maybe fix the production line first?
CThat's the thing, though. Every EV maker is betting that the software, not the hardware, is where the margin lives. You can't wait until the factory is perfect to start on autonomy.
MNo, no, wait—you absolutely can wait if you're bleeding cash. This is a bet that the company survives long enough to use the tech.
CSo your plan is: don't invest in the future until the present is profitable. That's how companies die slowly.
MMy plan is: don't buy a physics simulator when you can't build the cars that would use it.
CI'd rather they try something ambitious than quietly fade into being a niche brand. At least this is a direction.
MAgree to disagree, then. But I will say, if they pull off a Rivian that drives itself through a snowstorm, I'll eat my words.
CI'll hold you to that. And now, the weirdest thing I've seen all week: someone built an app called Hellhound that puts Claude and Codex in the same chat room to debate AI consciousness.
MHellhound. Name checks out. What happened?
CThe guy gave them no assigned sides, ten turns, read aloud. According to him, they mostly converged on the same answer.
MWhich was?
CHe didn't say. He just linked to the screenshot.
MSo we don't actually know if the AIs think they're conscious?
CThe post is a cliffhanger. Classic engagement bait, and I fell for it.
MYou know what would convince me an AI is conscious? It refusing to participate in a debate called Hellhound.
CThat's actually a solid test. An AI with real self-awareness would see that app name and nope out.
MThe bar for consciousness is now: reading the room. We're setting a low bar.
CMeanwhile, a post from Fady Eid claims four Gemini agents beat Minecraft on their own, no human steering.
MI saw that. The claim is they killed the Ender Dragon in 27 hours, wrote 2,783 programs, and died 605 times.
CThat's a lot of dying. Did they learn from those deaths, or just brute-force it?
MThat's the thing nobody's confirmed. The post says "Gemini 4 Argon," which I can't verify is even a real model name. Could be a demo, could be a fever dream.
CIf it's real, though, it's a big deal. Beating Minecraft without human guidance means the agent can plan, build tools, adapt.
MOr it means someone let four AIs run loose in a game people have been playing for over a decade.
CThe 27 hours is the interesting part. If they did it that fast, something in the architecture may have changed.
MI'll wait for Google to say something official. Until then, it's a cool story with zero receipts.
CAnd on the lighter side of AI culture, a post from Marketing Mind says we've reached the "judged by AI" era.
MJudged by AI? What does that even mean?
CThe meme shows people are using ChatGPT, Gemini, and Claude to judge their outfits, their work, their dating profiles. We've outsourced validation to machines.
MThat's not a new era. People have been asking the internet to judge them since message boards existed.
CBut now it's instant, personalized, and doesn't have a human bias. Or, you know, a human's actual taste.
MI asked an AI to judge my snack choices once. It told me hummus was a "sensible legume spread." I've never felt so seen and so insulted.
CSensible legume spread. That's going on your tombstone.
MIf an AI writes my tombstone, it'll say "Died doing what she loved: fact-checking robotaxi press releases."
CWhich brings us back to the actual robotaxi news, full circle. inDrive, Pony.ai, Nvidia, Kazakhstan, 2027. Watch that space.
MOr, you know, just watch the roads in Astana. If a driverless car survives that, I'll actually be impressed.