
La belle époque ττ.
8.1K posts

La belle époque ττ.
@RealGamersYTube
τhe golden age is yeτ τo come... τao


Training frontier models over the internet requires new techniques. Today, we present ResBM, a residual encoder-decoder bottleneck architecture that enables 128x activation compression for low-bandwidth distributed pipeline parallel training. Developed for @IOTA_SN9, we show SOTA compression without significant loss in convergence rates, increases in memory, or compute overhead. Expect the full paper release in the next 72 hours.

“The bomb he dropped at the end: - New algorithm for training "arbitrarily large" models on Bittensor - Targeting dropping in the next week or two - Positioning to go beyond what Templar did with dense 72B models - Jon: "to actually compete, like if you look at xAI, they're training 10 trillion parameter models now... we need to sort of cross that chasm... I think there's a way to do it” And just like that Templar will be just an old war story..



Arbos already runs a full subnet (SN97) with zero human involvement. The code is open-source. The weights are open-source. SN3 is still live. There’s nothing stopping a takeover. Headless Templar isn’t a theory. Everything needed to make it happen already exists.




"But Bittensor doesn't really do anything..." Most companies or crypto projects struggle to be the world's best at 1 thing. In just over a year, Bittensor has become the world's best at over a dozen things. And I've most likely missed a few subnets here. Bittensor is the Nasdaq of distributed decentralising AI... For me, it's literally a candy shop of some of the brightest minds out there building cool stuff. And all it takes is just 1 out of 128 subnets to achieve Product Market Fit and make hefty revenues and it will drag the price of TAO up. Because it will attract buyers into that subnet, but the only way to buy that subnet token is to first buy TAO and then swap it for that subnet token. Ethereum missed a trick here. ETH was never needed (other than gas fees) or linked to an ERC-20 project that mooned. With Bittensor, when you buy a subnet token, you're not selling TAO, you're temporarily swapping it...




🐋 WHALE WATCH:Grayscale is literally handing you the blueprint and people are still fading $TAO. They have allocated 43% of their AI Fund to Bittensor. Institutions dont ape based on hype they do deep due diligence. While you Are distracted by temporary FUD the big players are quietly cornering the supply of decentralized intelligence. The signal has never been clearer. Perspective is everything.



We think you should know exactly who's shipping your inference. No CEO. No board. No single person who can pull the plug. Ten contributors, a smart contract, and 100B+ tokens processed daily. @jon_durbin, Backend @airesearch12 (Florian), Backend + Research @KyleWidmann, Backend, built our TEE @0xsirouk (Chris), Backend @0xveight (Vince), Frontend @0xAlgowary (Timon), Sales & Customer Success @Fezicles_eth, Marketing & Community @RCSKBD (Rykorb), Technical Staff Cxmplex, Backend Vonkaiser, Technical Staff Our funds live in a smart contract that pays staking rewards to team members. No one person controls the money, the code, or the direction. Now you know. $TAO







