Subnet teams' own channels, X posts, press, newsletters and events read from the chain — GitHub lives in its own tab
Carbon@carbonphysicsaiThe most expensive design flaw is the one you find after you’ve built it. Physics AI helps engineers spot weak designs earlier, so they can spend more time testing the ones worth building. We want to put that capability into the hands of more engineers with the help of Bittensor bittensor:native
Carbon@carbonphysicsaiMost AI models are graded by the people who built them. Engineers want them tested on cases the builders never saw. Enter Carbon.
Carbon@carbonphysicsaiThese guys obviously haven't heard of Bittensor... “Physics AI can’t be trusted.” The exact problem the Bittensor mechanism enables us to solve. Detaching the builder from the evaluator driven by real incentives, real requirements, and real physics.
Quoting @StandardPhys
We are Introducing an AI engineer that turns physics simulations into better designs It redesigned a data centre’s cooling system to use less electricity and found a turbine blade that beat a leading published design Today we’re open-sourcing the simulation agent behind it
Carbon@carbonphysicsaiTestnet optimizing for mainnet to deliver SOTA
Quoting @carbonphysicsai
This forces valuable models only. This adversarial testing is how Carbon will deliver SOTA for specialized engineering problems. Incentive Computing will win. $TAO
Taosis News: the week’s articles from the subnet teams, the press and community newsletters, every Friday. Taosis Weekly: the week’s movers, best yields and chain events, with dollar values, every Monday. Past issues
Carbon@carbonphysicsaiHere's were we tell miners about incoming changes like these so they can prepare. Check it out! dashboard.carbonphysics.ai/#/updates
Linkdashboard.carbonphysics.aidashboard.carbonphysics.ai/#/updatesQuoting @carbonphysicsai
What did we learn this week? We tried to give the miners everything they could possibly need to build a great physics model... Practice data, training data, and the solver we used. But our exam didn't make sure that they actually NEEDED a great physics model. What does that mean? 🧵
Carbon@carbonphysicsaiWhy Engineers Use AI Alongside Simulation - Explainer Video
Quoting @carbonphysicsai
Carbon@carbonphysicsaiThe design competition is no longer multiple choice. Your model steers an automated design engineer toward the solution of a brand new hidden problem in the same problem space. Then your pick is evaluated against the best possible option the solver could find. No chance with a model fit to the data.
Carbon@carbonphysicsaiThis forces valuable models only. This adversarial testing is how Carbon will deliver SOTA for specialized engineering problems. Incentive Computing will win. $TAO
On Taosis
Swap TAO ⇄ USDC
USDC on Ethereum to TAO, or back. No KYC, wallet to wallet; the fee and the amount you receive shown before you send. Not instant: run by Taosis, paid by hand within hours.
Carbon@carbonphysicsaiValuable lesson learned before emissions turned on! Luckily. We can turn up the heat on this adversarial evaluation. Now, we give you enough data to learn the physics, but not to win the challenge.
Carbon@carbonphysicsaiHow? Now we are testing everything a data fit can't reach to force models to learn the real Physics. The accuracy cases we test are now... 1. Beyond range - Designs past anything in the training data. 2. New combinations - Input mixes that training data barely covers. 3. Regime switches - Where the physics changes behavior 4. Changed conditions - A different operating point on the same hardware. 5. Trends - Does the model get the direction and size of a change right. 6. Changed requirements - Pass/fail at shifted limits. 7. Joint corners - Several inputs pushed past their range at once.
Carbon@carbonphysicsaiOur exam was broken into 3 parts: 1. Physics gates. Fail and you're out. 2. Accuracy against the solver. 3. A design SELECTION competition against the solver. And every case we tested was new/hidden.
Carbon@carbonphysicsaiSounds great! What's the problem? Every case we tested was inside the range of data that the miners already had access to for practicing. So....miners dominated testnet by building models that fit the data the were given. Effective? Yes Learned Physics? No
Carbon@carbonphysicsaiWhat did we learn this week? We tried to give the miners everything they could possibly need to build a great physics model... Practice data, training data, and the solver we used. But our exam didn't make sure that they actually NEEDED a great physics model. What does that mean? 🧵
Carbon@carbonphysicsaiIndustry momentum 🔥 More on our investors page, link in comments.
Carbon@carbonphysicsaiInvestors: carbonphysics.ai/investors/
Linkcarbonphysics.aicarbonphysics.ai/investors/Carbon@carbonphysicsaiScore = Value. A subnet’s scoring determines what its miners build. We built ScoreLab to test whether higher scores lead to better engineering decisions. For each challenge, we compare scoring approaches against what matters to the customer: 🔸 Does the model respect the physics? 🔸 Is it accurate across the problem, including the hard cases? 🔸 Does it make the right decisions at the limits? 🔸 Do its recommended designs hold up when checked by the solver? 🔸 Is it fast enough to be useful? Physics and accuracy come first, speed can’t make up for a model that fails the fundamentals. The evidence helps us understand where a model can be trusted and where an engineer or the full solver should take over. Miners compete, Carbon learns from every attempt and customers get models built for the decisions that matter.
Carbon@carbonphysicsaiAn engineer is designing a cold plate for a new AI accelerator. Narrower coolant channels may remove more heat, but they also make the pump work harder. A design that controls chip temperature by consuming too much pumping power can undermine the efficiency you're trying to sell. We’re building Carbon so engineers can explore channel layouts together and take promising trade-offs into detailed testing.
Carbon@carbonphysicsaiThe cheaper it becomes to explore an idea, the more ideas get explored. Engineers can spend less time getting the first concept out and more time exploring what the best version of it is.
Quoting @AbelAIx
Picture an engineer with a drone drawn on a napkin Normally that takes hours of CAD work Skyfall's Sky 00-1 turns the sketch into a full 3D design, every part down to the screws They call it the first world model built for engineering > 100x quicker and 150x lower cost than GPT Astra, already ahead of the top labs Chatbots pay for every token, a model that pictures the job first could make AI cheap enough to build real things
Carbon@carbonphysicsaiWhat happens when a small team can explore ideas that used to need an entire department? AI is starting to put CAD, electronics, simulation and manufacturing tools into more hands. The next wave of hardware companies could be built by teams that look very different to the ones that came before them.
Quoting @OpenAI
ChatGPT and Codex are helping engineers build for the physical world. We’re sharing our progress across mechanical design, electrical engineering, and simulation—from helping our hardware teams iterate faster to testing new workflows with industry collaborators.
Carbon@carbonphysicsaiWe've had a lot to share with the Bittensor community and plenty more to come. On top of our own deep thinking about how to apply the Carbon model to Bittensor, in the lead up to @ExploitSummit, the @bitstarterAI team pushed us to get really clear on the what, how & why. Thanks for talking shop with @RyanBequette @mccrinbc & @macrozack
Carbon@carbonphysicsaiWe're optimizing Carbon testnet challenges daily in preparation for mainnnet. Updates are transparent, you can view fixes, know issues and questions here: dashboard.carbonphysics.ai/#/updates
Linkdashboard.carbonphysics.aidashboard.carbonphysics.ai/#/updatesQuoting @carbonphysicsai
Testnet is LIVE and man is it working! Miners are already crushing this. Building physics AI models, testing their approaches & competing across our four engineering challenges. We will spend this period working with them, and improving our exams so they are optimal for mainnet launch when real alpha is at stake! The dashboard is live, you can follow the results here: dashboard.carbonphysics.ai
Validator hotkey 5DvTpiniW9s3APmHRYn8FroUWyfnLtrsid5Mtn5EwMXHN2ed announced new child keys on SN6, SN8, SN9, SN14, SN20, SN29, SN30, SN35, SN37, SN38, SN41, SN44 and 30 more, applying at block 9,431,952 (~24h). Stake delegated to it there routes through its children after that.
Between blocks 9,424,713 and 9,424,772
Carbon@carbonphysicsaiThings are getting very interesting in the AI x engineering space 👀
Quoting @spisallyouneed
Very proud to showcase our latest demo of how our Engineering World Model SKY-001 can design a complex Jaguar car with 1000s of complex parts. In order to get us to Autonomous Manufacturing, we need to design systems that can think like a ‘AI General Engineer’: planning complex physical structures, long term sequences, and feasibility directly in the physical domain rather than approximating it through next-token prediction. Engineering design operates in the spatial reasoning domain through hierarchical decomposition, constraint satisfaction, and the sequential construction of complex physical objects. World models are going to be the next frontier in AI after LLMs and SKY-001 is just the beginning.
Carbon@carbonphysicsaiCarbon | Development Update: Oct 26 - Nov 2 Path to Mainnet 1) We Launched Carbon's Control Center! This was #1 on the list for me pre-mainnet. A full blown Physics AI research lab gated only by miner registration. We will be recruiting SciML researchers to come compete so we want an excellent front door. We want wallet set up > buy TAO > compete to be simple and fast. Then pick your Challenge, your compute, your inference, and your agent. All in one app. This is how we recruit elite talent and get 2500 miners building the future of Physics AI. 2) Testnet was next! Our first round of challenges got released to the miners on Saturday and they're smoking them 😅 Feedback has been great, and the scores are seriously impressive. Because of our design where miners submit model designs and we rebuild them, we are able to study each king and decide if we are getting the kind of models we really want! I think this is huge for our path to commercialization. What did we find? Two of the challenges were too easy! Miners nearly maxed out 1 of them already with really impressive strategies...but they aren't what we want them to be building. True physics learning models. These strategies were so good/unique that our internal testing didn't catch it....but now we know! The next round will have more inputs, harder hidden-cases, more complex design challenges, changing limits & boundary conditions. **ALL THESE CHANGES are being updated in the "Updates" tab of our dashboard so miners can prepare and know what's coming!** We're pushing until we find exactly how to get true physics learning models out of our challenges. Our design allows this, and this testnet period has helped us immensely! We won't stop until we do. By the time we are going to market, we will be certain of exactly what and how to test these challenges to get the highest value models out of them and we will have evidence. THAT is the path from now to revenue. 3) Mainnet launch is coming We're…
Carbon@carbonphysicsaiThere’s a wealth of knowledge among the Bittensor community and people backing Carbon. Carbon Council gives committed holders and key contributors a regular forum to share ideas and challenge our thinking. What should we prioritise? What are we missing? Where could we be more ambitious? 1000 alpha to join with no qualifying hold time for the first 30 days. Miners and key contributors by invitation. Check your eligibility and claim your seat via the link in comments.
Quoting @carbonphysicsai
👀 Sneak peek at something we're working on for our community: Carbon Council. The idea is to give our most committed holders and contributors a direct line to the team and a place to ask questions, share ideas and give feedback. We’ll host monthly online stakeholder meetings covering recent developments and the near-term roadmap, followed by open Q&A and discussion. Eligibility: → First 30 days after launch: hold 1,000 alpha. → After that: hold 1,000 alpha for more than 30 days, or 2,000 alpha for at least 14 days. We’ll also invite active miners and other valuable contributors directly. Several miners are already testing on testnet and giving Ryan useful feedback. Thank you. That’s exactly the kind of contribution we want to recognise through the Council. Once live, holders will be able to sign up with their email, connect a wallet to verify eligibility and claim their Council seat.
Carbon@carbonphysicsaiClaim your seat: council.carbonphysics.ai
Linkcouncil.carbonphysics.aicouncil.carbonphysics.aiCarbon@carbonphysicsaiJoin the Carbon Telegram community to learn more. t.me/carbonphysicsai
Linkt.met.me/carbonphysicsaiCarbon@carbonphysicsaiEvery car, plane and robot you’ve seen is the best design its engineers had time to simulate. Traditional simulation is slow and compute heavy, so most ideas never get tested. Physics AI models learn the physics from simulation data and predict how a new design will perform almost instantly, so engineers can explore far more designs before they commit to one. Carbon runs open competitions to build the best model for a company’s specific design problem, then tests every entry independently on cases it has never seen. The engineer gets a faster custom model and the evidence required to use it in real design work.
Carbon@carbonphysicsaiTestnet is LIVE and man is it working! Miners are already crushing this. Building physics AI models, testing their approaches & competing across our four engineering challenges. We will spend this period working with them, and improving our exams so they are optimal for mainnet launch when real alpha is at stake! The dashboard is live, you can follow the results here: dashboard.carbonphysics.ai
Linkdashboard.carbonphysics.aidashboard.carbonphysics.aiQuoted post
Carbon's Control Room is......open! A competitive Physics AI research lab on your desktop. 🚀 Beta test our launch challenges on practice exams now. We want zero friction between someone being interested in Carbon, and mining competitively on the subnet. This is our first step toward that. Testnet weights tomorrow. Find the bugs. Break things. We fix them. Then mainnet. Bring your expertise. Deploy your agent. Start building physics AI. bittensor:native
Carbon@carbonphysicsai"We're never going to stop building things" The TAM is big & we're confident Bittensor can help us address it well.
