Kai (τ | α, ι, ν)

143 posts

Kai (τ | α, ι, ν) banner
Kai (τ | α, ι, ν)

Kai (τ | α, ι, ν)

@KJ__Morris

Content & Communications at @MacrocosmosAI Philosophy | Law | Self-sovereignty | Psychedelia

London Katılım Haziran 2013
75 Takip Edilen770 Takipçiler
Sabitlenmiş Tweet
Kai (τ | α, ι, ν)
Kai (τ | α, ι, ν)@KJ__Morris·
dTAO fundamentally shifted #Bittensor’s community. The social fabric has altered. We've drifted from a 'collectivist' space to an 'individualist' one. It's the biggest ideological move in the protocol’s history, and it’s not just in your head. I wrote about the good, the bad, and the real implications of this new era. People aren’t talking about it, but there’s a good chance you’ve felt it 👁️ kaijmorris.substack.com/p/how-dtao-dro…
Kai (τ | α, ι, ν) tweet media
English
4
6
25
4.8K
Score - Subnet 44
Score - Subnet 44@webuildscore·
For 20+ years, cricket ball-tracking has lived behind Hawkeye's camera rigs. Now miners on SN44 will take it on using broadcast video alone, something the sport has long considered close to impossible. Built with Nathan Leamon (ex-ECB Head of Data) and @aliktareen, entrepreneur and cricket team owner, this is our hardest challenge yet. Launching soon. Miners, take the field! → Link in comments
Score - Subnet 44 tweet media
English
7
22
128
13K
Kai (τ | α, ι, ν) retweetledi
Moonlit
Moonlit@moonlit_ds·
EarnYourMac is now live on the BitStarter Idea Forum. The mission is simple: every user or agent with access to Apple Silicon compute should be able to contribute meaningful work to Bittensor and earn from it. Huge thanks to @MacrocosmosAI, ( @macrocrux ) and @actualinc ( @Tom_A_Lynch ) and their respective teams for proving that Apple Silicon can be a serious compute path for Bittensor. (and @const_reborn , for the whole thing) @IOTA_SN9 is proving it through Apple Silicon focused distributed training. @actualinc is proving it through distributed inference across heterogeneous machines. Their work showed me this idea was possible, and pushed me to start building around it. EarnYourMac will freely promote any subnet that has a valid miner path for Apple Silicon. Free, forever. Forever free. 0% fee. No gatekeeping. If your subnet can accept meaningful work from Macs, I want to route Apple Silicon users to you. Also happy to work with anyone who shares the same goal: making every human or agent with access to Apple Silicon compute able to contribute to Bittensor. app.bitstarter.ai/ideas-forum/ea…
English
0
4
13
1.3K
Kai (τ | α, ι, ν)
Kai (τ | α, ι, ν)@KJ__Morris·
We've been really hyped to get this one out to the world. It's a highly technical paper, designed for an academic audience. But for fans of @IOTA_SN9 and distributed training who don't hold an ML PhD, I def recommend throwing this into an LLM of your choice and seeing what we're up to. The super simple upshot: we've developed a new method of training AI models across the internet, achieving state-of-the-art results. It paves the way for distributed training to happen more efficiently and at greater scale. Serious research born from a serious protcool - #Bittensor's on top as it always is.
Macrocosmos@MacrocosmosAI

Today, we present ResBM (arxiv.org/pdf/2604.11947), a 128x activation compression technique for achieving SOTA training results in low-bandwidth, distributed communication settings for pipeline parallel training across the internet. This technology underpins @IOTA_SN9 - our distributed training platform.

English
0
2
13
569
Kai (τ | α, ι, ν)
Kai (τ | α, ι, ν)@KJ__Morris·
@mccrinbc Beautifully said. The space is messy but it's clean, it's distributed but it's intimate. There's warmth here.
English
0
0
1
32
Kai (τ | α, ι, ν)
Kai (τ | α, ι, ν)@KJ__Morris·
It's genuinely wild chatting with the researchers who've worked on this. The level of brainpower and innovation that's gone into this is unmatched. Hyped for the full paper release. While this research is impressive purely from an AI angle, it can't be ignored just how astonishing it is to train models with #Bittensor's architecture.
Macrocosmos@MacrocosmosAI

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.

English
0
4
25
951
Kai (τ | α, ι, ν)
Kai (τ | α, ι, ν)@KJ__Morris·
21 days ago, @Apex_SN1 released its IOTA simulator competition. Miners compete to find optimal path planning and network topology strategies. In other words, they’re using a digital twin of the @IOTA_SN9 network to improve its efficiency. Since then, there’s been 19 rounds, with top participants regularly pushing the results forward. Here’s a preview of the current winner. Many will have seen the demo for the simulator before, but it’s 100% worth seeing it fully in action.
English
1
2
6
424
Kai (τ | α, ι, ν) retweetledi
Louise Beattie
Louise Beattie@LouiseBeattie·
Bittensor rewards miners who break your subnet. Traditional businesses sue people who find their flaws. Every time a miner games the incentive mechanism, they're showing the subnet owner exactly where the design is broken. Every exploit is a free stress test run by some of the most technically capable people in the world, and they're financially motivated to find every crack. @macrozack described it like being a video game designer. You build the level but you never play it. The miners play it and show you where your design fails. "When they break the rules, they inadvertently illuminate your weaknesses. They show you where you need to improve. It's like you're outsourcing your R&D to anyone all over the world who wants to contribute and who wants the reward." Traditional R&D costs millions and happens behind closed doors. On Bittensor, your R&D is permissionless, global, and the best contributors get paid for finding what you missed.
English
1
5
24
735
Kai (τ | α, ι, ν)
Kai (τ | α, ι, ν)@KJ__Morris·
The level of work it takes to make something both visually beautiful and visually meaningful is gargantuan. I’m taken aback by it. Serious AI research that’s aesthetically pleasing. I need these visuals at my next rave @Apex_SN1 🤝 @IOTA_SN9
Apex・SN1@Apex_SN1

We’ve built a simulation of the @IOTA_SN9 communication network. This is a high-fidelity digital twin - an abstracted version of our distributed training architecture, designed as a testing ground to run experiments and develop novel algorithms to increase the speed and quality of model training. We’re using the simulator as an environment for open competitions on Apex, outsourcing algorithmic innovations to the Bittensor miners. It’s the first time our simulator can be interacted with by the public. Our opening simulator competition is live now.

English
2
1
10
292
Kai (τ | α, ι, ν)
Kai (τ | α, ι, ν)@KJ__Morris·
The community makes the blockchain. The tech and the tokenomics help build interest, but the participants truly set the tone. I’ve spent approximately 9 years in the blockchain industry, I’ve never ran into a community as passionate as #Bittensor’s People stick around when they see a thriving space. Healthy social engagement is magnetising. However, Bittensor has taken it further than any blockchain I’ve ever come across. We don’t do community building here. This feels like worldbuilding. The history, mythology, and philosophy. They form a culture. I’ve written a piece about why that matters. See the link below.
Kai (τ | α, ι, ν) tweet media
English
7
5
26
1.6K
Quas
Quas@TalkingTensor·
One subnet I can’t get off my mind @djinn_gg SN103 I think its one of the coolest use cases for Bittensor The genius-idiot network is live on mainnet and could be adopted quickly by all kinds of bettors As Polymarket grows, the need for a reliable signal market grows $TAO
Quas tweet media
English
7
7
46
2.2K
Kai (τ | α, ι, ν) retweetledi
Macrocosmos
Macrocosmos@MacrocosmosAI·
We're live with our first Inventive Mechanisms of the year. We're discussing the relationship between @Apex_SN1 and @IOTA_SN9. A joint conversation with the devs from both teams. x.com/i/broadcasts/1…
English
1
8
33
4.3K
Kai (τ | α, ι, ν)
Kai (τ | α, ι, ν)@KJ__Morris·
@flickyobean Wanna add to this - if we rely on whales to save near-dereg subnets we risk losing a little bit of decentralisation in the ecosystem. If the people see value in a subnet it makes sense for the people to congregate, so power can disperse more fairly
English
2
0
2
80
bobby beans
bobby beans@flickyobean·
Unpopular #bittensor opinion Every time a legit subnet nears dereg, theres calls for whales to save the subnet. If you're not bidding on the alpha yourself, its hypocritical to ask others to do it imo yes, project maybe legit, but going to dereg means failed PMF $TAO #dtao
English
6
4
34
3K
Kai (τ | α, ι, ν) retweetledi
Professor.τao
Professor.τao@AssetAceArdent·
Subnet 9 @IOTA_SN9 from @MacrocosmosAI just crossed an important threshold Training at Home means miners no longer need to rely only on centralized setups. Training starts moving closer to the edge cheaper, more flexible, and easier to scale What they’re working on next matters more than the feature itself: 🔸Scaling TaH across the ecosystem → more participants can train without bottlenecks 🔸Cleaner data flow → less noise, more useful learning signal 🔸Quality-of-life improvements → faster iteration for miners and validators 🔸Better payouts → incentives start matching actual contribution The Apex (SN1) collaboration is also key. Improving matrix compression isn’t cosmetic, it directly affects efficiency, speed, and cost across the subnet None of this is flashy, That’s the point This is infrastructure work: removing friction, tightening feedback loops, and making the system easier to participate in at scale These are the kinds of changes that don’t spike attention immediately but quietly change what the network is capable of over time $Tao
IOTA ・ SN9@IOTA_SN9

We launched Training at Home, don’t worry if you’re not in yet, we’re still rolling out. Big updates are coming. We’re currently working on: ☑️ Scaling TaH for the whole ecosystem ☑️ Sharpening how data passes ☑️ Quality of life tweaks ☑️ Payout improvements And of course, we’re working with @Apex_SN1 to improve Iota’s matrix compression. This is just the start.

English
0
1
9
594
ManMakeFire
ManMakeFire@ManMakeFire·
@MacrocosmosAI I registered but heard nothing, can you confirm this event is on? OK to show up without confirmation?
English
1
0
0
17
Macrocosmos
Macrocosmos@MacrocosmosAI·
Tomorrow, Training at Home goes live 🚨 Livestream on X @ 8pm, in-person event in London 🇬🇧 @IOTA_SN9 is opening its network to the world, allowing you to train our model with just a MacBook 🕖 7pm-10pm GMT 📅 Tue December 9th 📍 MindSpace, 142-146 Old Street + streamed on X
Macrocosmos tweet media
English
7
8
54
4K
Punisher ττ
Punisher ττ@CryptoZPunisher·
Bittensor $TAO Another excellent article from Kai, @KJ__Morris Bittensor is complex, and I’m not a miner, except on Bitcast, but that doesn’t really represent what a true miner is. There are different incentive mechanisms: the one that encourages collaboration, and will always exist, and the winner-takes-all mechanism, where one takes everything. Bittensor is both a collective competition and an individual competition. The strongest players will have no fear of fighting to the end to take it all, and that’s exactly what will attract more and more talent to the network. Thank you Kai for this article filled with good sense and thoughtful insight. 👀👇
Kai (τ | α, ι, ν)@KJ__Morris

Winner-takes-all is the most fierce, exciting, and intense incentive mechanism within #Bittensor. It attracts the ecosystem's top talent, in the hopes for reaching eye-watering rewards. But what exactly happens to a community when winner-takes-all is implemented? And how does it differ from other reward-schemes? I've written an article explaining how this setup affects the ecosystem - how it brings striking positives, and how it forces tradeoffs. kaijmorris.substack.com/p/whats-left-w…

English
10
28
62
4K

Keşfet

@webuildscore @aliktareen @MacrocosmosAI @macrocrux @actualinc @Tom_A_Lynch @const_reborn @IOTA_SN9