I would challenge that a bit. I'm not pushing the analysis forward to real use cases, I'm making the analysis more complete. Your basically looking a cost/token, right? But I'm saying that there is no such thing as a language agnostic token, so you must think at the level of language. Ex: if you compared a Deepseek's revenue to OpenAI's, you must first convert RMB to USD or USD to RMB, right? Theoretically, if you get deep enough into a model's layers, you can directly interact with its latent space, where you can achieve a sort of language-agnostic token (or a language that is native to the LLM). In fact, Looped LMs think at this level (instead of tokenspace). But at the level of the input/output, the LLM cannot be divorced from language. 所以你必须得考虑语言

Any low-key bangers I should know? 👀💎 No hype, just real plays 👇 #HiddenCoins #ConvictionPlays #CryptoResearch #TokenSpace #NextPump

Any low-key bangers I should know? 👀💎 No hype, just real plays 👇 #HiddenCoins #ConvictionPlays #CryptoResearch #TokenSpace #NextPump

gradients flowing through tokenspace

But how does this work, exactly? We've used an array of self-hosted, distributed, and decentralized technologies to make it a reality: •︎ The WhiteMarbleDAO is an Urbit Syndicate; its membership and operations orient around the ~pondeg Urbit identity (instantiated as an ERC721 NFT on mainnet Ethereum). •︎ ~pondeg is one of 65,535 possible syndicate token creators in a 'finite tokenspace' implemented by the SyndicateRegistry contracts, preventing a deluge of token creation spam and scams. •︎ ~pondeg created the $WHTMRBL token, which grants membership to the DAO with initial membership and funding coordinating via a crowdfund using %fund (a self-hosted crowdfunding tool). •︎ ~pondeg runs an instance of Urbit OS, which hosts the portal.whitemarble.ai front end, an instance of the %privateer chat interface, granting access into the White Marble DAO Library. •︎ The chat interface connects across urbit's p2p encrypted network to an Urbit 'moon' (subservient digital identity controlled by ~pondeg) running on a separate machine which runs the %computeer software, connecting to an @AnythingLLM that holds the contents of the library and turns it into a private RAG. •︎ This private instance of AnythingLLM then connects to a distributed set of Mac Minis which get tasked with inference jobs coordinated by @GradientHQ's Parallax Software.

I really want to know how they managed to train a model which can output: Steer/Accel, Language, 3D Gaussians, ? Is it just trained Tokenspace? How did they get the language in there? How did they get sub 1/4s 3D gaussians in there!?

We are thrilled to Announce the Partnership with -@Tokens_Space TokenSpace specializes in securing CEX listings and delivering targeted marketing solutions for blockchain projects. #CEXListing #partnership #LBankEDGE #Bangladesh

i wonder what would happen if you embodied an LLM in a tokenspace world. a nonstop stream of side-channel observation tokens flowing into the context with the ability for the agent to move around as it talks, constantly running during the conversation with the user

Word analogies are cool because it’s just testing how calibrated your internal tokenspace is

" %slab is a token launchpad for the "finite tokenspace" that also includes a tokenbound account wallet for your @urbit identity. send and receive crypto directly between self-sovereign digital identities on a p2p network " - @tocwex

%fund lets you do p2p crowdfunding. %slab is a token launchpad for the syndicate ecosystem's "finite tokenspace" that also includes a tokenbound account wallet for your urbit identity. send and receive crypto directly between self-sovereign digital identities on a p2p network

for (layer 1) planets, it serves as a tokenbound account interface for your urbit id. stars and galaxies can do quite a bit more, though. namely, launching a 'Syndicate Token' into the finite tokenspace (~65k possible erc20 tokens). right now it is a closed beta, available only to stars which have booted their instance of urbit os and are operating on urbit's p2p and e2ee network. if you want in, get on the network and shoot me a dm, and/or join ~tocwex/syndicate-public.

I mean this just goes back to the OG thesis that LLMs are persona simulators, no? to write evil code it gets into the "headspace" (tokenspace -- what's the difference?) of an evil person, and the rest follows the SAE jockeys will tell us soon enough I reckon

gunna keep repeating myself. finite address space. finite namespace. finite tokenspace.

i'll say it again... finite address space... finite namespace... finite tokenspace...

Wow, that's some serious hodling power in the black hole there! Elon's taking over the tokenspace one lock at a time #DogelonArny

This particular prompt does alot of reprinting (much of the tokenspace for each iteration contains the growing context)... I've only done a few tests but I got up to around 5 iterations before it stopped output. I wasn't so much impressed with the prompt as I was w/ the model

Thinking often about this plot in the wake of #o1. Whether CoT resembles real human reasoning is irrelevant. The leading labs don't want to mimic human reasoning, they want to surpass it. "Reasoning" in tokenspace is brilliant for alignment, as long as output is human readable.

Lots of extra ways you can mutate at the root. Especially tokenspace permutations and corpus permutations.

ve been working on this project for years and bought TokenSpace but I have never claimed it from the site. Of my own accord. That's interesting @SpaceTokenBSC @CorpDutchy @janrost4 @royasamani1 @humanmani1