AZKABAN

I’ve got 10 GTD spots for you guys. @sporefolks A 3,333 collection powered by $SEED, where holders apparently gotta use Seeds to mint and upgrade their Specimens. Mint goes live on our beloved Robinhood: ⟣ mint is tomorrow, 3 PM UTC ⟣ Mp starts at 6,666 $SEED ⟣ Mechanism: Uniswap V4 To enter: > Follow @sporefolks > Rtw this post & Drop your EVM wallet below ⏳ 7HRs

I’m starting to think robotics is about to change what the AI race even means. For the last few years i think it’s been pretty easy to follow. ⟣ Who has the smartest model? ⟣ Who reasons & codes better? ⟣ Who wins the benchmarks this week. But robotics is slowly making that race a lot messier. Because now, the models have to actually understand the world and do shit inside it. Plus the market is starting to accelerate fast. ⟣ XPeng’s robotics arm just raised $900M+ at a $6.3B+ valuation. ⟣ Skild raised $1.4B, & Generalist raised another $200M barely two months after a $400M round. So yeah, there’s a stupid amount of money moving into physical AI rn. But the money is honestly the less interesting part to me. I mostly care more about what everyone is trying to build with it. Because I think the race is slowly moving beyond who can build the best robot body. Into who can build the intelligence inside it And we’re already getting hints of what that stack might look like. Gemini Robotics is one example: They’ve been using higher level reasoning alongside lower level control models. And companies like Figure’s Helix splits slower reasoning from the much faster system actually controlling the robot. Even companies like Anthropic have been experimenting with how useful Claude becomes when it can give instructions to a pretrained robotic policy. And this makes way more sense to me than stuffing the biggest frontier model possible inside every humanoid and asking it to control every finger 50 times a second lol. But I also think frontier models eventually sit higher up the stack. Maybe they become the teacher. They try to understand the environment, solve the harder problems, generate data, and transfer some of that intelligence into smaller embodied models that can actually run locally without murdering latency or battery life. And If that’s the direction we’re slowly moving on to, then the AI race could get a lot more interesting. Because then we start to question: > Who owns the body? > Who owns the brain? > And who owns the model the brain’s been built on? The robots themselves also eventually become data generators. Meaning every object they pick up, or a failed attempt they encounter can potentially become training data for the next generation of embodied models. And that creates a pretty insane loop. frontier model teaches robot, >> robot goes into the world >> robot generates physical data >> that data trains better models >> better models get deployed into more robots. And repeat. I believe we’ve spent the last few years watching labs fight over internet scale text, code, images & video. Maybe, and just maybe Physical AI could create an entirely new internet scale dataset.. imo that might end up being one of the biggest moats in robotics. PS: China is insanely good at this robotics thing atm. I just watched a video of a robot beating usain bolt’s PR.