Openτensor Foundaτion

This week on Novelty search :: Mechanics of Conviction - Locks decay by default. - Subnet owner emissions can be auto-locked - Short, clear unlock periods. - Exponential conviction maturity. Join @const_reborn tonight, live via Bittensor Discord

Bittensor is getting its own confidential routing layer :: @say_gm_ Simple payments, use for any AI model. Miners compete to serve inference at the best price. “What our miners do is they provide the tokens at the cost that they decide, but they’re competing against other miners…” @mogmachine + the team building #SN28 is bringing the OpenRouter structure into Bittensor: routing, payments and model access as part of the subnet economy. SN28 is live on Testnet. NS075 full episode :: youtu.be/eQYvR0yZaTg

Brand new real-time subnets screener page released on taoflows.app/subnets 👀 Dexscreener style. Selected global timeframes. Sort by Flows, Volume, Price %, Emissions, Code score, etc. Discovering Bittensor $TAO subnets made easy

Bittensor Ecosystem Highlights of the Week #63 // SUBNET UPDATES & ACHIEVEMENTS ➤ @chutes_ai SN64 Jon Durbin successfully demonstrated decentralized training of a 176B-parameter model across four internet-connected nodes using Parallax. (bit.ly/4ecgYYd) Chutes rebuilt its infrastructure, significantly reducing model saturation and improving reliability across its flagship models. (bit.ly/49vTXwO) ➤ @webuildscore SN44 They have deployed their first vision agent in production on an Avia fuel station. (bit.ly/4dPwo3h) ➤ @zeussubnet SN18 Zeus launched their new rebuilt website. You can try their weather forecasting API without a key on it. (x.com/zeussubnet/sta…) (x.com/wouterhar/stat…) ➤ @heydittoai SN118 DittoCode is live, you can now build and share apps on Ditto. (x.com/heydittoai/sta…) ➤ @SwarmSubnet SN124 Swarm developed a proof of concept of a real-time fire detection from an autonomous drone, using @manakoai to detect the fire and @chutes_ai to decide what to do. (x.com/SwarmSubnet/st…) They’re expanding beyond individual drone challenges with a new robotics lab, broader benchmarks, and a focus on general autonomous flying agents. (x.com/SwarmSubnet/st…) ➤ @actualinc SN95 Actual Client v0.14.0 is now live for private beta users, with a major UI update across macOS, Linux, and Windows. (x.com/Tom_A_Lynch/st…) ➤ @TargonCompute SN4 Over the past month, they have repurchased 2000 TAO of SN4 alpha using organic revenues. (x.com/TargonCompute/…) ➤ @theminos_ai SN107 Minos expanded its live variant-calling benchmark to chromosome 21. (x.com/theminos_ai/st…) Minos will be presenting two works at The American Society of Human Genetics (ASHG) this fall. (x.com/theminos_ai/st…) ➤ @QuasarModels SN24 They dropped their Quasar-Preview model on Hugging Face. (x.com/QuasarModels/s…) ➤ @404gen_ SN17 404 showcased an early agent-built browser game using procedurally generated Three.js assets from their upcoming competition. (x.com/404gen_/status…) ➤ @desearch_ai SN22 x Ditto SN118 Ditto agents are integrating Desearch as their native search layer. (x.com/desearch_ai/st…) ➤ @QuantumSN48 SN48 Half-price quantum compute is live. (x.com/QuantumSN48/st…) ➤ @yanez__ai SN54 Yanez is showcasing Proof of Humanhood and Proof of Uniqueness as a way to prevent fraud without exposing identity data. (x.com/yanez__ai/stat…) ➤ @b1m_ai SN105 Beam teasing Beam Transfer Studio, a platform for orchestrating distributed data transfers across clouds, databases, and storage providers. (x.com/b1m_ai/status/…) ➤ @Ninja_Subnet SN66 Ninja shipped a POLAR-style rollout pipeline to turn live coding-agent competitions into training data. (x.com/Ninja_Subnet/s…) ➤ @trishoolai SN23 Trishool’s Halo improved its F1 score from 75% to 87%, closing the gap with QwenGuard. (x.com/xnavkumar/stat…) They also shared their updated product roadmap. (x.com/trishoolai/sta…) ➤ @affine_io SN120 They’re invited by @Alibaba_Qwen to attend the very first global Qwen Conference in Singapore. (x.com/affine_io/stat…) ➤ @ai_detection SN32 They released a Canvas LMS plugin, to bring AI-generated content detection directly into schools and universities' existing workflows. (x.com/ai_detection/s…) ➤ @mvtrx_79 SN79 GenTRX is live on mainnet. (x.com/mvtrx_79/statu…) // SUBNET INVESTMENT ➤ @IOTA_SN9 SN9 @stillcorecap has invested in SN9 alpha token. (x.com/markjeffrey/st…) // NEW SUBNET ➤ @say_gm_ SN28 The Venice of Bittensor is here. (x.com/mogmachine/sta…) ➤ @Taolepathy SN25 @KeithSingery finally decided to end the embryon competition and use his slot to develop his own subnet. (x.com/KeithSingery/s…) // BITTENSOR ECOSYSTEM ➤ @opentensor Conviction upgrade is now live on mainnet. (x.com/bloomberg_seth…) // PODCASTS & ARTICLES ➤ @opentensor Novelty Search with @const_reborn on the mechanics of Conviction (x.com/opentensor/sta…) ➤ Hash Rate by @markjeffrey with @centrum_blue from @theminos_ai (x.com/markjeffrey/st…) ➤ @VenturaLabs podcast with @peytonspencer and @sebyrubino (x.com/VenturaLabs/st…) ➤ Ventura Labs podcast with @Swamination from @YumaGroup (x.com/VenturaLabs/st…) ➤ Inventive Mechanisms podcast with @macrocrux and @Austin_Aligned on @Apex_SN1 x @AureliusAligned (x.com/MacrocosmosAI/…) ➤ @SubnetSummerTAO AMA with @babelbit (x.com/babelbit/statu…) ➤ Subnet Summer AMA with @vocence_bt (x.com/SubnetSummerTA…) $TAO

A few weeks ago, almost everyone hitting our flagship models got a 429. Utilization was pinned at 95–100%. Demand had outrun our compute. So we trimmed the catalog to the models people use and pushed that compute where the traffic goes. Most flagships have headroom now. A few still run hot, and the autoscaler is scaling them. Don't take our word for it. Our utilization dashboard is public and updates every minute: chutes.ai/app/research/u… What should we measure next?

Limited edition: the Bittensor wallet, powered by Tangem. @opentensor @bittensor Cold storage for $TAO, in card form. Pre-orders open now. bittensor.tangem.com

Bittensor’s Conviction upgrade just hit mainnet today. The feature has a large scope and is the most wide reaching upgrade since Dynamic TAO, impacting every stakeholder on the network. Important note before diving into all the details: as of today, Conviction is functionally opt-in for subnet owners and has no material impact on stakeholders (e.g., subnet owners are at no risk of losing their subnet). Subnet owners and the community are being given time to explore the mechanics of the upgrade, discuss what settings work best for both tokenholders and subnet operators, and can/should voice feedback during this initial phase. Those within the network know Bittensor protocol development moves fast, so below you’ll find everything you need to know about the upgrade and how it’s being rolled out. Summary of Conviction Conviction is an onchain token-locking primitive. Users can lock alpha tokens to a key which (1) creates a "conviction" score and (2) locks tokens for a certain period of time. The intent is for subnet owners to be able to lock up tokens to signal their long-term conviction for building on the subnet. Eventually, functionality will be added that enables anyone to contest ownership of a subnet (they’ll need to have at least 10% of the Alpha held by tokenholders backing them), and if successful (i.e., a new key has a higher conviction score than the owner key), take control over it. Subnets that are less than 1 year old will be excluded from this feature. This functionality is not yet enabled on mainnet. Locked tokens can be unlocked by calling an onchain function. The unlock schedule follows an exponential curve, with 50% of the locked tokens available after 3 months, ~95% after 12 months. Locked tokens can still be transferred for OTC investments, but they will remain in the state they’re received in (e.g., if an investor receives locked tokens they’ll remain locked unless they choose to unlock them). For now, Conviction is primarily a social signaling feature. No subnets can be taken over. Subnet stakeholders will all need to work together to decide what the norms/expectations will be for each subnet, as they all have different constraints. What’s on mainnet, how it works, and what’s not on mainnet A user (any coldkey; subnet owner, regular token holder, etc) can lock some amount of their alpha on a subnet to a specific hotkey. While locked, that alpha cannot be sold. Over time, the lock also generates a "conviction" score that follows a continuous exponential formula. When locking, the user can choose between two lock modes: • Decaying lock (the default): locked alpha exponentially unlocks over time. After 3 months, half is liquid; after one year, ~95% is liquid. • Perpetual lock: locked alpha stays locked forever (until the user toggles to decaying). A given coldkey can have at most one active lock per subnet. Top-ups on locks must target the same hotkey (no locking to two different keys). Moving a lock to a different hotkey is allowed but resets conviction to zero. Subnet owners can decide whether their owner emissions are auto-locked or liquid. If set to auto-lock, every owner emission for that subnet is automatically locked to their hotkey rather than flowing in as liquid stake. The default on mainnet is liquid, not auto-locked. Subnet owners effectively have to opt-in to Conviction at this level. Subnet owners will also have to decide whether to lock all, a portion, or none of their current holdings, and whether a lock would be set to decay or perpetually locked. For a simple example: under a decaying lock with these defaults listed above, locked tokens and conviction evolve as: • Days Elapsed = 30; % still locked = 79%; conviction score (% of locked tokens) = 18% • Days Elapsed = 90; % still locked = 50%; conviction score (% of locked tokens) = 35% • Days Elapsed = 130; % still locked = 37%; conviction score (% of locked tokens) = 37% (peak) • Days Elapsed = 180; % still locked = 25%; conviction score (% of locked tokens) = 35% • Days Elapsed = 365; % still locked = 5%; conviction score (% of locked tokens) = 17% Under a perpetual lock, locked tokens stay constant and conviction matures to its asymptote on the same 90-day half-life curve. The system is not symmetric across all users. Subnet owner conviction is immediate. When the subnet owner's own coldkey locks alpha to their subnet, conviction is set to equal the locked token mass instantly. There is no 90-day ramp up. The owner has the most to commit and gets credited for it immediately. Anyone locked to the subnet owner's hotkey is also granted immediate conviction. Subnet owners have a structural advantage in keeping ownership of their subnet. Transfers carry locks with them. Often, teams will execute OTC deals with investors via token transfers; these will still function the same way as they do today. However, when alpha moves between coldkeys, the lock follows the alpha. The receiving coldkey inherits the lock state and can choose to make it perpetual or immediately decay. So, if a subnet owner transfers an investor locked tokens, it will be up to the investor to decide when to unlock them. The unlocking event is also an onchain event. The hotkey association on token transfer must match the receiving key. If the receiving coldkey has an existing lock to a different hotkey, the transfer fails. This prevents locking across different hotkey commitments. Right now, Conviction scores are computed, stored, exposed via RPC, but nothing onchain consumes them. The one function that would consume aggregate conviction, change_subnet_owner_if_needed, which would replace a subnet's owner with the highest-conviction hotkey's coldkey under certain conditions, is not functional yet. Think of what’s shipped today as the initial, slow rollout of the grander Conviction design. What to expect now All subnet stakeholders should start understanding, experimenting with, and discussing the design and parameters of Conviction. Every subnet operator has different funding constraints, trust levels, and maturity on a relative basis. A large part of Conviction will be settling on the norms and expectations the community/social fabric has with respect to subnet teams. Many of these parameters discussed above can be tweaked via governance. The takeover design/mechanism has yet to be rolled out yet, so it can still be evolved based on community feedback. Come join the community call/Novelty Search tonight to hear Const discuss these changes and provide feedback directly.

Just linked my OpenClaw to my @heydittoai account. Ditto now knows everything my OpenClaw knows. They share memory. They're a hivemind now. Holy crap.

Back from the first global Qwen Conference in Singapore. The most valuable part was the technical time with the @Alibaba_Qwen model team: 1. Distillation for compact models using highly filter data, teacher logic/KL alignment and curriculum design 2.Coding-agent post-training and long-horizon SWE environments 3. eval and training pipeline infrastructure. We left with sharper questions, better intuitions, and several ideas to bring back into Affine’s SWE-frontier env and Distill env design. More soon.

DittoCode is LIVE! Build and share apps with the world! Now the only constraint is your imagination. Next, you will be able to monetization your creations to anyone on the internet. We can't wait to see what you make!

We’re live on our Inventive Mechanisms podcast. @macrocrux and @Austin_Aligned are discussing our upcoming competition. This is a collaborative task with SN37, @AureliusAligned, launching on @Apex_SN1. Join to learn more x.com/i/broadcasts/1…

The greatest minds in TAO. One room. The Louvre. The Bittensor Track is back at @proofoftalk for year two, and we're co-hosting the #TensorFi workshop with @wearetalisman. @mikecontango, @contangojosh, @kenjon of @bitmind, @philism of @forevermoney_ai, and @wearetalisman will be discussing: 🔹 What can DeFi unlock for the Bittensor network — liquidity, investment primitives, market making, the agentic economy? 🔹 How can Bittensor's AI infrastructure improve and evolve existing DeFi protocols? June 3. Paris. Register here: luma.com/glytiyi4

TEE routed private inference: saygm.com/?ref=c5a239b34…

Proxy accounts are live in Talisman. If you've done a $TAO limit buy/sell or invested in a TAO managed index, you've probably already used one without realizing it. Now you can see them / revoke them / add your own. Here's why that matters 👇

Be at @MuseeLouvre with $TAO 🔥🇫🇷 Having subnet owners, validators, allocators, researchers, and builders all in one place at the same time feels pretty significant for the ecosystem. That’s why the dedicated #bittensor track could genuinely end up being THE highlight of @proofoftalk. Personally, I’m especially looking forward to: ⚪️ Subnet Keynote by @shardiban (@oroagents) ⚪️ Proof of Pitch: Live Crowdfunding with @bitstarterAI ⚪️ Fireside chat between @const_reborn and @ninabambysheva on “The Unification of AI and Bitcoin” 👀 See you there! 🍷🥖

Next week, @bittensor in Paris :: @proofoftalk

Proof of concept: real-time fire detection from an autonomous drone Live camera → fire detection → annotated evidence → agent decision loop → movement command @manakoai detects the fire @chutes_ai decides the next move @SwarmSubnet turns it into a flying agent We're not ready for what happens when subnet intelligence enters the physical world Code tomorrow. Open-source. Again.

At the Qwen Conference Singapore, we joined the AI x Web3 panel to share Bittensor’s incentive design, Affine’s post-training methodology, and the upcoming Affine Agent. Great discussions with teams across the industry including represents @coinbase @LBank_Exchange @0G_labs on how AI is reshaping Web3. The Bittensor ecosystem and Affine are clearly at the frontier of decentralized AI. Next post⬇️: takeaways from our conversations with @Alibaba_Qwen model researchers 👀

Just for clarity: This has the potential to fundamentally disrupt both the current AI compute market and the frontier LLM training race. Why? Because Parallax by Chutes is starting to demonstrate something that was widely considered impractical until now: large-scale decentralized training for frontier style MoE models. If this continues to scale (we have reasons to believe it will), the implications are huge: - frontier training across massively distributed hardware - significantly more cost efficient to train - has the potential to free up huge load of the severely constrained data center capacity, that is currently tied up for centralized training (estimates say 20-50% of dc compute is locked for training) - heterogeneous training fleets, e.g. high-end NVIDIA B300 GPUs alongside consumer hardware like MacBooks - the possibility for ordinary people to contribute compute, participate in training frontier AI, and earn crypto rewards with it, basically: models from the people for the people. Yes, this is basically SETI@Home for frontier model training. Democratizing the benefits of AI. AKA one of the core visions of Bittensor as I understood it. /cc @const_reborn , please correct me if wrong. And the most insane part actually comes AFTER training: The same architecture choices also appear to massively improve inference efficiency. Think multiple times better revenue/cost ratio for inference providers. Oh, and btw, Chutes is an inference provider. Most people still underestimate how big this could become. Stay tuned. Or no, even better: Please share. We need to get the message out there. Except if you are an investor or want to train a huge model, in which case: please reach out via DM. $TAO