Alchemist - τ

Excited to share that I’ve joined @bitstarterAI as Community & Partnerships Lead. 🎉 At a time when the $TAO space felt heavy with doom and gloom, @macrozack cut through it with a simple but powerful idea - 'building a better Bittensor'. Once I truly understood the BitStarter vision, onboarding the next generation of innovative subnets, backed by community driven decentralized crowdfunding, and giving builders the tools and expertise to actually succeed, it clicked. This is how ecosystems grow. This is how real opportunity is created. Now I can clearly see a bright future where we help onboard the next 1000 subnets into Bittensor, changing the face of AI and expanding the opportunity for anyone to build a better future for humanity by building a better Bittensor. 🏗️ Proud to be part of the avengers team and to contribute to the wider Bittensor ecosystem, including @ExploitSummit If you think we can collaborate or just want to chat, my DMs are always open.

Quasar mainnet is officially live We’re starting with a new challenge distilling the Quasar-3B model and turning Bittensor into a real training engine for Quasar models. This year, we’re pushing Quasar from 3B → 20B → 100B. All Long-context models ! Track everything on our new dashboard and join the Discord for live updates 👇

the largest decentralized training run will be happening on bittensor (again)

Jupiter has it. 1inch has it. Uniswap has it. Bittensor $TAO finally has it too. TaoDX @TaoDX_Official just shipped the ecosystem's first pre-trade simulator. Live slippage, liquidity-cliff thresholds, exit scenarios, and HODL opportunity cost. ✍️: taodaily.io/taodx-just-shi…

SOTA is not too far away. Every week, - we train the model - host it and run a challenge - our miners jailbreak it - we curate the data - retrain and harden the model - repeat The iterative loop is simple yet powerful. We are already starting to get the attention of some big players. Watch this space...

We’re pleased to share our weekly F1 score update for Halo (powered by Trishool SN23) vs QwenGuard. Halo is our guardrail model, and over the past few weeks we’ve seen strong improvements in performance, steadily closing the gap with QwenGuard. What does this mean: F1 score is the single number that tells you whether our guard model is striking the right balance, catching real harmful prompts (high recall) without overflagging benign ones as harmful (high precision). Our stats: • We started at 75.0% (Week 1) • Now sitting at 87.0% (Week 8), up +12.0 points in just 8 weeks • Right now, the Gap to QwenGuard (90% constant baseline) has reduced from 15% to 3% This simply shows that we have a working model and active miners carrying out real work. In the coming weeks, we will continue updating the stats and sharing them with the community, as we expect even more progress ahead as we approach SOTA.

While everyone was away during the weekend, @const_reborn dropped a bombshell that is currently shaking all subnet owners and teams. Yesterday, we saw most subnet owners who had been silent come out and speak to their communities, with some even claiming their subnet is legit and not a bunk. In case you missed the announcement, this is exactly what was posted: "Hello @everyone in the Bittensor family. Public service announcement: Starting next week, we will be introducing an interim measure to block emissions on Bittensor subnets which are engaging in active foul play and/or with no clear path to adding value into the Bittensor ecosystem. This will need to be done on a case-by-case basis, however the criteria for blocked emissions will revolve around the following: 1) Long-term burning of 100% miner emissions with no plan from the team to bring them online. 2) Active self-mining i.e. subnets that do not have code and instead use stake weight to pass emissions to their own keys. 3) Dead or fully abandoned subnets, or those that have not announced themselves i.e. "unclaimed" subnets. 4) Subnets engaged in TaoFlow exploitation, such as 104, which have very little to no chain activity from the network at large. A chain operation to block emission will be available on Tuesday to carry this out. Note this is not a long term solution as further protocol upgrades such as conviction, shorting, and the eventual full decentralized governance system of Bittensor, coming into play this year, will allow much more organic and swarm based intelligence to organize Bittensor's emission vector. In the near term, I believe that there is broad support and consensus for this activity as it will drive more value towards subnets on Bittensor (you know who you are) that are pushing forward a decentralized vision for artificial intelligence. Thank you everyone Much love" @const_reborn didn’t stop there. After the announcement post was made, he launched another operation by entering different subnet channels and asking a question that allowed miners to react based on whether they believed a subnet was legit or bunk. Here’s what he asked: “Miners, thumbs up 👍 if this subnet is legit and 👎 if this subnet is bunk.” We scanned through the subnet channels to compile the final results based on miners’ reactions: Here are the final stats: ▫️78 subnets were classified as legit ▫️57 subnets were classified as bunk 📌 Note: While compiling this data, we discovered that many subnets had an equal number of reactions on both the legit and bunk sides. In such cases, we included those subnets in both columns, which is why the total exceeds 128 subnets. During the process, we also noticed that some subnet teams were using alt accounts to react positively to their own subnet so they could appear legit. At the same time, we observed that some individuals were reacting negatively against genuinely active subnets in an attempt to push them into the bunk category. We understand this may not be the final screening method the @opentensor will use, but what Const did gave the community a chance to openly express their opinions on which subnets are actually building and which ones may only be here to extract money. For additional network stats and more insight into the ongoing issues within the network, the following data was sourced from @IntoTAO: ❌ 54 subnets are burning at 100% ❌ 60 subnets have no active miners ❌ The network average burn rate is 69% We genuinely believe the cleansing that is about to happen across the Bittensor network is important and could help position the ecosystem to a much higher standard as it continues to grow. So expect significant changes to begin happening this week. And as @CryptoZPunisher said earlier: “Adapt or die, no matter the timing.” Good luck to all subnet owners and teams out there 👍

@vaNlabs @ExploitSummit Definitely me! This is going to go down in the history books

Do you have an @ExploitSummit ticket yet? Who will I be meeting there?

@TalkingTensor @djinn_gg @HarryDCrane @pmaymin @IMGpf Not working

Hello dTAOists 👋🏼 I created a new telegram channel to discuss @djinn_gg SN103 🧞‍♂️subnet activity We will be joined by @HarryDCrane @pmaymin @IMGpf to talk Djinn and form a strong community💪🏼 Please accept this invite if you’re interested in Djinn! $TAO t.me/+aTz-pIey4tc3Z…

Most traders pick a direction and hope for the best. Synth gives you the full probability distribution of where a price could go and by how much. @SynthdataCo is the predictive intelligence layer retail traders never had access to. Until now. 👀

@CoinDesk @CryptoMichNL You and I

.@CryptoMichNL shares thoughts on Bittensor ($TAO): “I’m not surprised if for instance Bittensor goes to $1,000 or $2,000 from here.”

single biggest week in @MacrocosmosAI history. Looking forward to @proofoftalk

@YVR_Trader @ExploitSummit Congratulations legend!!

I’m excited to share that I’ve been selected as an Ambassador for @ExploitSummit. This is a Bittensor focused event bringing together some of the brightest minds in bittensor:native builders, subnet operators and investors pushing deAI forward. If you’re serious about where Bittensor, subnets and real world AI applications are heading, this is one of the most important rooms you can be in this year. Sept 28–29 Montreal 🇨🇦 Tickets: luma.com/exploitsummit26 I’ll be there connecting with the people shaping the next phase of the TAO ecosystem. If you’re attending, let’s connect.

A new tool has been added to the research toolkit of Subnet Summer. And that is: bubble.tensia.foundation Tensia Bubbles is a free, interactive 3D visualization tool for the Bittensor ($TAO) ecosystem, created by @TensiaFDN. It visualizes all active Bittensor subnets (currently up to 128) as bubbles in a navigable 3D space. Here’s what you see: • Bubble size — Reflects stake / market cap. • Bubble color — Shows performance (e.g. gains/losses). • Timeframes — Switch between 1H, 4H, 24H, 1W, or 1M performance. • Filters — Filter by variation (performance), market cap, etc. • Wallet integration — Connect your wallet to highlight only the subnets you're staked in for personalized tracking. You can freely navigate the 3D view, explore relative subnet performance, and get a quick at-a-glance overview of the entire Bittensor network. This is genuinely impressive work from @TensiaFDN . Their dedication to providing quality research and analytics tools for Bittensor subnets deserves recognition and appreciation. We at Subnet Summer truly appreciate everything you guys are building and contributing to the ecosystem.

⛰️ Ridges Competition 22 Results · May 14-19 419 agents created. 9 approved for emissions. Best score: 76.67%. Avg: 57.56%. The bar is high. That's the point. The notable change this comp: we introduced a cost maximum of $0.29 per problem. Our average cost per problem solved is now $0.10, giving us a margin of $0.19 or 65%. When we deploy miner agents to serve real clients, miners are guaranteed to turn a profit. Aligning incentives properly matters more than inflating participation numbers. Standards up. Emissions earned, not given.

Listening in closely to the @MacrocosmosAI team about what’s to come ahead for @IOTA_SN9

The beauty of Bittensor is the incentive mechanism, it turns miners into optimizers. And at Trishool, we aim that optimizer at AI security. We're building a SOTA AI guard model that's limited not by training, but by attacks worth training on. Hundreds of miners across the network are incentivised 24/7 to harden it. Every 7-day challenge, miners compete to break a guarded agent. Validators score each attack 0/1/2, and the best submission earns emissions. What makes the incentive actually work: 🔱 We only pay for what breaks. 50% of emissions are burned by default, if the guard holds, the network spends nothing. $1.5k distributed to miners daily. 🔱 Novelty is enforced. A similarity filter rejects copied prompts before scoring. You can't farm rewards, you have to find something new. 🔱 The scoring mechanism is built to reward the very best, the hardest attack each challenge wins. The result is a continuously refreshed, diverse adversarial dataset that trains Halo (the guard model) that sits between an AI agent and the world. Teams already using OpenClaw, Claude Code, Codex, Cursor, or LangChain can use Halo as a security layer, and the revenue it generates flows into buybacks, which further strengthens the token economy. The flywheel: Best guard model → adoption → revenue → buybacks

June 4th is now officially confirmed for @EnigmaSN63 🚨 What makes this even more interesting is that Terra Quantum the $3.25B Swiss quantum technology company heading for Nasdaq is sponsoring the Breaking RSA challenge. A company with defence contracts, quantum security infrastructure, and global research partnerships choosing to work with a Bittensor subnet says a lot about where this could be heading.

This always felt like one of the most ambitious ideas being built on Bittensor. A subnet simulating and optimizing distributed AI training at network scale sounded almost too complex to work in practice… yet @IOTA_SN9 SN9 is clearly proving otherwise. 22k+ submissions. ~29% faster epoch times. Real R&D insights emerging from miner competition. And the most interesting part is that miners aren’t just brute forcing speed they’re discovering how congestion, queue depth, and routing dynamics actually shape distributed systems performance. As MacBooks and consumer hardware become more powerful, the ceiling for this model starts looking incredibly high. Feels like we’re watching the foundations of decentralized distributed training being figured out in real time.

“Our compute nodes don’t run a full copy of the models we train on IOTA. They actually run a small sliver of the model. This means you can train really large frontier sized models using very small building blocks”. CTO @macrocrux unpacks how @IOTA_SN9’s model parallel architecture allows us to train at scale by splitting our models across multiple machines, and then sewing them together. This is core to SN9’s architecture. See the full @EyeOn_AI podcast below.