1/ SCALING THROUGH DECENTRALIZATION This piece explores the motivations behind improving model performance in a decentralized setting. Link in next tweet.

1/ Running Bittensor. Myself and @derekedws co-wrote this AI x crypto piece to describe the core innovations of Bittensor. medium.com/collab-currency/r…

@dylan522p (leading AI researcher) was asked about what spaces in the AI startup world he is is excited about. his first answer? distributed training and inference. in particular, @NousResearch and @PrimeIntellect

1/ NFT value comes from attention. However, attention has a ceiling if sufficient liquidity does not exist. Attention= Value, but Attention x Liquidity= Value^2 Check out my piece on capital efficiency and NFT liquidity solutions, summary below. medium.com/collab-currency/m…

1/ "Ethereum's future staking market dynamics” This post takes a focused view on the overlooked yet important factors facing the decentralization of Ethereum's staking layer. medium.com/@ronanbroadhead/1…

I get to compare the @believeapp flywheel to Ubers 'Virtuous Cycle' described by Bill Gurley, because everyone is abandoning crypto tokens for crypto equity proxies im sorry but these are just the rules

annihilation of meme mania allows for aeons to thrive in my mind, separates wheat from chaff….mrkt proved who’s in it for long haul

medium.com/@ronanbroadhead/a… Check out mí post about @AnataNFT and VTubers. @bigdsenpai @dcfgod @CapitalGrug

“i always knew decentralized training was gonna be big, yup mhm. crazy actually like a year ago i was telling JR that something about these ‘communication bottlenecks’ didn’t sit right with me” loading t-90d

this interview was from 2017 so with a grain of salt but, sacks: - speaks of @coincenter in high regard - believes in protocol coins (non-security tokens) - believes in security tokens, cited @sbmckeon's tokenization framework not my first choice but🤷‍♂️ medium.com/hackernoon/tradit…

1/ @machibigbrother’s $APE proposal is not narrowly scoped to its end goal, and must be denied. While the proliferation of Yuga assets in the traditional art world is a worthy task for the DAO, this proposal is neither rational nor efficient.

ofc the @Citrini7 optics blog dropping the same day as the @NousResearch DisTrO report: Citrini: LLMs are going to require a massive connectivity build out cycle Nous: our initial training run shows a 857x bandwidth reduction in pre-training, and 10000x for post-training / fine-tuning....oh and larger models may have greater reduction for my dreamers: - dePIN is necessary to properly incentivize the connectivity build out - distributed/decentralized can more efficiently use existing infra, and reduce energy requirements

@Citrini7 is a gud writer tldr; -inference on device (edge) enables personalized AI (users won't be comfy giving their data to big tech) -transforms AI from a generalized understanding of the world to a generalized understanding of *you*, will give rise to more specialized fine-tuned models (model of models) and be the driver of new consumer behaviors and AI apps -apple uniquely positioned to benefit: existing distribution + user trust + ability to vertically integrate the ai stack (good at hardware and software) = largest iPhone replacement cycle in decade+ -phone manufacturer supply chain benefits as a result (not subbed can't see tickers) citriniresearch.com/p/themat…

if you follow me at all, you know just how bullish i am on @PrimeIntellect couldn’t ask for a better team to go after the biggest opportunity in tech

if you believe in the ETH_is_money thesis, then the tcos buying it is only part 1, part 2 is them slamming it into defi

Updates: -Recently started (full-time) on the @Collab_Currency investment team. -Very recently moved to San Fransisco for the beautiful weather. Lmk where to send my application for the SF crypto cabal. would actually love to meet anyone down here :) picture unrelated

🕺💃

there’s maybe ~20ish people who are properly bullish decentralized training, and its viability just doubled up in the reasoning paradigm it is arguably blockchains’ largest and most consequential use case, lock in.

incredibly excited to be backing @xponentcrisis and @ckcheung in building Tensorplex. Few have the necessary lens on both crypto and AI to push τhese products. stay tuned for their upcoming subnet launch :)

one of the coolest things that’s happened this year combining savy crypto participants (the compute contributors) with cracked AI team (@PrimeIntellect) this combo is going to lead to the biggest outcomes in tech imho

1/If crypto is programmable money, and @delegatecash is a programmable proxy, then Liquid Delegates are programmable escrow. This thread tries to outline the power of Liquid Delegates (LDs) and some novel applications of them.

I will keep repeating myself b/c somehow this is still going under discussed: teams leveraging decentralization to compete with the major labs in the biggest tech race in multiple decades, are actively moving up to the adult table

recalibrate all previous levels of bullishness on @PlasmaFDN higher, effective immediately karen’s a killer and shes gonna crush it🤍

Just had to give my Uber driver a 0 ⭐️ because he told me threw 98 and ran a 4.3 with a straight face.

full stop best dTAO interface is @BackpropFinance by @TensorplexLabs terminal for charts, recent trades, top holders, liquidity and valuations, screener for discovery, and native swapping FRESH

'I think new entrants will get much more excited about the yield crypto can generate through liquid restaking and if we're lucky, gaming items' wow we are so lucky @onchaingaias

led the investment group at @oregonblock, which helped get me an internship and eventual FT position at C+C stoked to see this go live, and serves as a good reminder that the incoming workforce grew up mostly online with no predispositions about what can and cannot hold value 🍜

its novel b/c it incentivizes intelligence through gamification and speculation its like if mario kart was used to train AI models with an embedded casino to incentivize better intelligence, but the players owned the AI and the casino

got an X-Ray on my foot that showed a bone fragment from my heel-- doctor wanted me to come in for an MRI b/c they didn't know if it was from an old or recent fracture put images into Deep Research and it concluded the fracture was recent -- logic was very consistent, and well-cited. Even citing a study that showed an 88% missed detection rate for the type of fracture it identified, with a bunch of ancillary evidence that it was indeed a fracture p crazy

*memes pumping* every perps guy ever: floki good for a punt here imo

I personally am very excited by the prospect of net new entrepreneurs building real businesses, not memecoins, using crypto!

Your pfp community is posting phising links or virtue signaling, ours is discussing maths and EV outcomes We are not the same @jebus911

mom calls me today: ‘the oligarch tech right billionaires can manipulate these models to their own agenda!” the more of this grok like incidents, the more open source + dTraining enters the zeitgeist

wen @believeapp hackathon with $1M prize pool

asking chat a question for it to source you, admittedly feels kinda cool

gm jumbled friday thoughts a primary driver of markets today is the ongoing capEX cycle into AI, and all of the downstream funding and buildout. equity multiples and late stage private valuations are increasingly rich, making investors uneasy. the current capEX trajectory also may be vulnerable if the ROI the AI complex is seeing as a whole doesn’t pick up. Importantly, ROI is the concern - lots of big ARRs with similarly large compute costs OpenAI is in many ways the bellweather of AI. If they can produce meaningful ROI, it can support the argument for further investment across the board. 3 weeks ago they launched GPT5 and with it a router to understand the intent of user requests. Meaning, the router can decide whether a prompt is economically monetizeable or not, and spend more compute on high value prompts. Clearly moving to a model where they can serve a transaction from discovery to purchase, and charge a take rate. (h/t @SemiAnalysis_) This moves their business model off of just [raw compute<>intelligence output], and shifts their TAM to all of online purchasing. A lot of work is still needed to make the agentic checkouts and plug-ins better, but it’s a big deal b/c it can prove alternate ways to monetize (in turn proving ROI). Also caveat that OpenAI finding better ways to monetize doesn’t necessarily mean other players will be able to, but it may push them to continue investing or finding new ways to produce +ROI.

Wassie x Anata: The NFT you can literally KILL Thanks @rootshe11 :)

'when people ask me what my GTM strategy is, I say crypto....the ace up my sleeve is this community' 'why launch on believe' b/c all the believe apps site visits and users 4x'd overnight nitter.app/i/spaces/1yoJMoVBEawKQ

"Venture Capitalists Are Funding a New Wave of Crypto Mania on College Campuses" shoutout @zackrosenblatt_ and your not bullish enough on @Dorm_DAO

4/ The promise of decentralized AI has never been stronger, particularly as we enter the reasoning paradigm. If you are interested in leveraging decentralization primitives to compete with the big labs, please reach out.

2/ Proper understanding of previous and new scaling paradigms is critical to understand the role of decentralization in scaling AI. Can decentralized AI compete? Why do we want, or need, to create and serve AI decentralized? medium.com/collab-currency/s…

absolute scenes

originated from @404gen_ minted natively on Bittensor EVM @0000ALCI

2/ We believe that Bittensor represents the natural counter-balance to the existential problems that centralized AI poses today. In this piece, we explore: -Bittensor x Bitcoin parallels -The state of AI -Bittensor's Unique Architecture -Future Use Cases / Applications

very clear that the number of apps/startups is about to accelerate with AI tooling advancements, similar to e-commerce and Shopify stores go look at the landing pages of every YC Spirng 25 company, they all look the same

if you would, please consult the graph no further questions at this time

funny that my favorite two brands in crypto (@PrimeIntellect + @NousResearch) are both d/ai

>bets against myself >goes AFK

"Even with increased delay (up to 4 steps), Prime-RL matches the performance of synchronous baselines."

5/ Thank you to my co-author @derekedws, + amazing insight/feedback from @Shaughnessy119, @shibshib89, @chrisF_0x, @jameswoodmanv, @theemozilla, @xponentcrisis, @ckcheung, @ai, @jasminervaa, @KeithSingery, @benroy__, @jackbutcher, @Jerryshixd, @mrjasonchoi, @toptickcrypto, @CarlKVogel, James at @I_D_Theory, @atleykins, @andrewjiang, @Old_Samster + my team @collab_currency.

public equities: liquid private late stage companies: semi liquid private early stage companies: illiquid if tokenization is largely a means to create more efficient and liquid markets, then tokenizing YC startups should be much more impactful than tokenizing OpenAI or TSLA

very excited to get this out, should be a good one :)

@notthreadguy is basically the Kai Cenat of CT

this research from @FelixLts addresses many of the 'known unknowns' that the @LidoFinance Staking Router modules introduce. it lays the groundwork for a standardized and comparable module framework: -Fee structure -Operator registry -Capacity for validators -Internal distribution mechanism -Slashing coverage -Performance and decentralization metrics -Module differentiation it analyzes: -the potential effects of the Staking Router on Ethereum's overall stake distribution -the practical constraints of module capacity + transitioning to a new stake distribution model across modules: and conceptualizes a future stake distribution flow given said constraints: Felix introduces 4 main possible distribution mechanisms: -Governance based -Dynamic tokenomic based (gauge mechanism) -Algorithmic based -Hybrid based (combo(s) of the above 3) and discusses the relevant tradeoffs of each mechanism + introduces a goal based evaluation framework: -Module innovation and contributor diversity -Decentralization and ecosystem alignment -Operator performance and stETH APY -Minimize slashing and other risks -Enable stakeholder participation in decision making This research opens the door for Lido to reach its goals of 5,000+ node operators, and reach new markets in the form of Restaking (through Eigenlayer delegation or natively) and institutionally-focused modules. as you can see, the devil is in the details. Lido must account for the relative preferences of stETH holders, LDO holders, module creators, while maintaining its decentralization and censorship resistant properties AND sufficiently minimize risk for all parties. Not to mention the technical and implementation complexity with evolving variables. shoutout @FelixLts and co.👏👏

yessir bittensor is completely untethered from its original foundation (OTF) it feels like the velocity of money (native network asset) is significantly faster without a centralized foundation raises, grants, M&A, talent, marketing -- all bottlenecked by one big entity

‘a market based approach to incentivize the right behaviors so that the right actors are more successful’ okie

Why fooo

idk if we have fully internalized what a 'crypto is now legal' moment truly means the bid is stronger when you believe in the tech

3/ I cover the following topics to answer these questions: - Current CapEx cycle - Scaling Laws in Pre-Training & Cluster Efficiency - Pre-Training Methodology - Distributed Training Approaches - Reasoning Models and Inference-Time-Scaling - Decentralized Networks

3/ Simply, Bittensor enables the creation of intelligence markets via cryptonomic primitives, directing resources to those best equipped to supply them. It serves as a coordination layer, granting developers full autonomy in the design of such incentive mechanisms.

Be me: Graduate college today Walk on stage and receive diploma placeholder IT IS AN ADVERTISEMENT TO PAY FOR THE ALUMNI ASSOCIATION Nice👍🏻👍🏻

their ‘forced narrative’ our big bet let’s see

ok so if im interpreting this correctly: - distilling works - reasoning models require sig less communication - current rung on the scale ladder requires fair bit of compute but not multi billion scale ILL TAKE DECENTRALIZED FOR 1000, ALEX

The consensus view is that decentralized AI will be useful for a number of use cases, but not for model creation. I wrote this piece to show why this view is misguided, and how decentralized AI actually provides the ultimate avenue for OS AI to compete with the major labs.

the ETH ETF narrative is a far more interesting conversation than the BTC one given there are real 2nd order effects due to staking. i cover these effects in this post, with guidance and input from @adcv_ 👑 medium.com/@ronanbroadhead/1…

*bunch of actual scams go up some then down a lot* yup cryptos a scam, 9th inning, industry is dead get a grip!

“i don’t see an inflow which has happened yet which makes it seem very premature to specifically ask the question is the cycle over as it relates to altcoins”

dancing!

When your mom asks u to clean your room but you're trying to be creative. #PSAT

Our Season 1 teams for the Bittensor hackathon have selected their projects, check them out! if you are interested in connecting please reach out

CK and @xponentcrisis sit in this unique position where they actually understand crypto native behaviors/levers, and are day 1 builders in the bittensor eco dangerous

1/ Product/Service vs Store of Value Go check out @derekedws new piece to be digital-objects pilled.

catch me pouring drinks next week in Marfa! There won’t be a tip jar, but will be accepting avant-garde digital art alpha🤠

On a train and sat down next to an older gentleman We talked about the fourth turning, AI, generational differences, prescription drugs, natural medicines, crypto, and how overwhelmingly optimistic he is ab my gen and the future Wrote this in my notes a year ago We so up🙏🏻

have worked with a lot of these students over the past year+ some of the idea generation and analysis on par with industry standard more to come, let @zackrosenblatt2 cook🍜

much conversation recently about whether or not ‘ETH is AI money’ wrong framing imo, trillion $ question is which asset can achieve the monetary premium status associated with the intelligence derived from open systems and co-ownership

P: what is the fringey part of crypto that you are interested in?’ M: if AI is intelligence, crypto is a way of coordinating intelligences. most are focused on agents paying each other….an interesting step further is crypto as ways to construct markets and allocate resources.

if you told me 6 months ago that Ethena would have 3x the FDV of LayerZero, i would have called you crazy

. -R2 will be 40% cheaper than R1 -deliveries slated for first half next year -infamous headlights that give it personality -‘i bought this tesla before i knew elon was crazy’ sticker phenomenon -amazon + VW relationship plugged_in

ok

gm we have 8 cracked teams from top universities with legit experience both in ai and crypto if you are a subnet operator, validator, miner, or holder who has something they would like to see get built in line with the tracks listed in this thread - hit me or @zackrosenblatt2

im assembling a team

hodge podge of catalysts: - the other labs drop their reasoning approaches + grok 3 bangs -> agi accepted as t-24months - SOL/dinos ETF - staked ETH ETF - USG commitment to not sell seized btc - crypto specific legislation, namely market structure bill - crypto fund LP appetite up -> smash and raise - quality AI networks launch - moonshot/trump onboards new participants - trump lip service for pro growth - dollar and rates at maxima BROS

gm

4/ In time, proper ownership structures are critical for a truly 'open AI' world. Read the full piece on medium here. medium.com/collab-currency/r…

will be here and would love to meet peeps

20 mins into the debate I found out that me and Hilary relatively have the same haircut......

jumping in front of a waymo to improve its training data diversity accelerate.

Would be crazy if the whole altcoin cycle was just a Decred fractal😭

🤓

latent space is like an abstract map of patterns a model learned Alci's NFTs (3D point clouds) have always reminded me of these latent space visualizers I like to think of them as representations of the latent space of Bittensor and @404gen_

kaito is blowing up my low follower gems smh

bunch of push_the_boundries shit has launched this year, particularly in the 2nd half but crypto's shot at the big labs hasn't even left the sling yet bros

i’m hooked on blackbird at this point restaurants have not had a venue that allows them to push incentives for growth in a customizable manner Blackbird is Pendle for restaurants, stay with me now

9/@Proteccxyz: A dedicated ecosystem partner and NFT market maker. The inefficiencies in the NFT market offers the opportunity for a “delta-neutral” market marker to provide significant value to NFT collections while capturing substantial yield.

anything is possible >4k ETH >5k and you can bring back the metaverse and gaming

casual Sam banger, highly recommend: "To achieve this, we’d want to build a framework to allow for anyone to contribute to a massive library of different reasoning samples (called “traces”) for a diverse set of tasks. Contributors should be able to not only submit traces, but also to create new environments to generate different types of data in a standardized way. That is, we’d want standardized environments for generating traces across math reasoning, physics, medicine, engineering, writing, and more. A robust spectrum of environments to generate and collect these traces would lead to a massive database for anyone to tap into for fine-tuning." this quote encapsulates a very important advantage that centralized AI cannot compete on, and why the reasoning paradigm is such a tailwind for dAI. we know crypto can unlock compute/capital, but the labs already have that. what if it unlocks the collective intelligence of a burgeoning research area? the labs do not have that.