i used to think prediction markets were mostly about waiting to see if you were right. but looking at @punchdotwin, i think the more interesting part is what happens after you take the position. most prediction markets make you wait for the answer. you form a thesis, take a position, then wait for the event to resolve. PUNCH is trying to make that feel more like trading. you find something you have a view on, choose the market, take a position, then decide whether to manage it, exit or let it settle. the underlying can be btc, gold, stocks, or even a football matchup. so the question shifts from “what do i think will happen?” to “do i want to put a position behind that view?” the one-minute markets make this especially clear. instead of waiting hours or days to test your thesis, you can try again almost immediately. what also stood out to me is that PUNCH isn’t just putting a prediction interface on top of it. there’s a CLOB orderbook, market and limit orders, and $0.001 price increments. makers pay 0% fees and can receive rebates of up to 50% of the taker fee when their orders fill. that part makes it feel much closer to an actual trading venue than a simple bet slip. but the real challenge is still liquidity. creating markets for everything is easy. building enough activity for people to trade those markets comfortably is harder. if PUNCH solves that, would you still think of it as a prediction market, or just another place to trade your view? explore more here: punch.win

good morning frens check out the echobit creators pogram if you haven’t already. i sent my application, now let’s see what they have in store for us. in the meantime enjoy my weak ahh game play. 😂 do have a great day ❤️

okay i just spent way too much time on fal.live it’s basically an endless stream of AI-generated content and somehow you keep finding another thing to watch. “infinite brainrot” might actually be the best way to describe it, lmao. @fal

what if the problem with robot data isn’t actually the amount of data? 5m trajectories sounds crazy, but there’s a bigger question: how much of that data is actually helping the model learn something new? if a robot already knows how to do something, giving it thousands more examples of the same behaviour probably isn’t solving much. the useful data might be the stuff around the failures. the tasks the model keeps getting wrong. the movements it struggles to reproduce. the situations where it looks fine in training but breaks down during evaluation. that’s the part i find interesting about @axisrobotics. the goal isn’t just to keep making the dataset bigger. it’s figuring out what the model is still bad at, then using that to decide what data should come next. because for physical ai, more data and better data aren’t always the same thing. explore more here: s.kaito.ai/NhyCbg2