
Building neurons, turning the web into a brain. Missionary from the Church of RAO @opentensor
Joined February 2023
- Tweets3.3K
- Following466
- Followers31.2K
- Likes7.9K
Proof of Useful work so useful it creates state of the art models from latent GPUs and talent networks.
Cascade MoE has beaten Moirai2 - a leading forecasting model by Salesforce.
Cascade scores 0.511 - a 0.005 improvement on the GIFT-EVAL CRPS benchmark - assessing probabilistic forecast quality.
Cascade MoE is also now at 86.2% parity with Timestral (@MistralAI's frontier model) on MASE - assessing point-forecast accuracy.
The upshot: we're getting closer to creating leading forecasters directly within #Bittensor, accessible to all.
We’re moving fast. This competition was only launched 19 days ago.
See the dashboard: dashboard.cascadesub.net/sta…
Your wildest game ideas can be brought to life with the 404 game generator.
This is Veilbound, an original anime sword adventure built entirely with the 404 game recipe (Path B code geometry, no imported 3D assets).
Every environment and character is created with Three.js code, from the seven playable worlds to the sculpted human fighters.
You don't need Blender or Unreal Engine to create a game of your own, just the 404 game recipe and an AI agent.
Welcome to a new era of game creation, powered by 404.
Try it out: github.com/404-Repo/404-game…
and btw, we can't do any of this stuff properly in Web2.
Crypto will never die.
const retweeted
that's why
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
images.nvidia.com/pdf/Open-W…
OSS AI is just a catch all.
What we need is permissionless compute, non KYC inference, encrypted model storage, decentralized pretraining, and collaborative RL.
Super intelligence is either co-created and co-owned with free access or it will become cancerous and asymbiotic with humanity.
We will turn the web into a brain. We will decentralize. We will be permissionless. We will be free.
Democrats have a wishlist for AI that includes a federal permit to train a model, a Department of AI, 20-year prison terms for executives, kill switches, and ID checks to use a chatbot. Writing code is about to need a hall pass if certain politicians get their way.
reclaimthenet.org/democrats-…
Yesterday we released rolling hash wallets (like Bitcoin) and Quantum protected wallets.
They combine together: you can wrap a quantum wallet under rolling hashes, multi-sig it, and use a vault signer, with a transaction delay.
You'd need to crack both Sha256 and quantum encryption -- at the same time -- to touch them.
Gamma and Quantum resist wallets are live on Bittensor.
Release notes:
bittensor.com/releases/v487-…
bittensor:native
Proud to see the release of REAX 1.3.
Its the worlds most performant jev-style decision model (independently benchmarked by @huggingface).
It tops leaderboards -- its was also trained by anonymous engineers on the internet through a mining network.
huggingface.co/spaces/multim…
We've been perfecting the art of training models in decentralized networks @bittensor.
They now move faster than companies at solving hard problems because they source international talent without borders, and we only pay for improvement.
Bias free and permissionless is also a great, ethical deal for the world's talent. You get paid directly for the work you do and in the ownership of the network company you helped create.
Bravo to the team behind this release, who went from zero to world SOTA in less than 2 weeks.
xitter.cf/reaxlabs/status/211870…
🥇 REAX 1.3 is #1 on the Hugging Face Decision Index.
We produced the best decision model in the world.
274 models. One top spot. Built on Bittensor $TAO.
Four weeks ago, Subnet 92 didn't exist. Then the miners went to work: an open competition, two selected miner models, one post-production run.
Now REAX 1.3 sits at the top of Decision Index 0.4.1 with a Full score of 74.0.
This is what decentralized AI looks like when it ships.⚡
Live today on a Jev-compatible API
decisionmodels.io
huggingface.co/spaces/multim…
Another Bittensor release which shouldnt go silently. This is a full top to bottom model, pre-trained on Teutonic (3) and then post trained in a competitive environment to reach coding performance above Qwen's flag ship 9B model.
These models are small sure, put these performance metrics are not, and the thing to watch here is alignment -- the team was able to aim a laser cannon at intelligence production and punch through SOTA weight class.
Every day something new. Congrats @reliquary_ai
Forgot to mention, they also used Bittensor compute @lium_io (competitively sourced from anons) and Bittensor sandboxes @TargonCompute (also competitively sourced from anons). Connectionism go Brrrrr. Bravo.
This is the second subnet this week that has made physical world touch down (The other was @metanova_labs's connection to a automated drug lab)
This bridges the gap between the liquid intelligence layers on Bittensor, which exist purely in byte space, down into the hard atomic space where our systems can optimize and innovate harder products.
Very cool leap.
We've signed our first commercial robotics data pilot with Vulcan Shield Hungary, one of the largest advanced material manufacturing complex in Europe.
Starting at a production workstation in Békéscsaba, we'll combine real-world data and simulation to work toward a robot model trained for a specific factory task.
const retweeted
✅ BENCHMARKED: IOTA Decide 4B v0.4, an open-weight decision model from @MacrocosmosAI, running on my 64GB M5 Pro Mac mini
You hand it a state (a ticket, a policy, an agent trace) plus questions and one forward pass returns a probability for every answer. With their bundled loader on the Mac's GPU in full bf16, it loaded in 12.8 seconds and peaked at 10.7 GB.
Short ticket, about 130 tokens:
- yes/no: 88 ms
- 3-way choice: 105 ms
- 3-level score: 88 ms
That's about 11 decisions a second on a desktop Mac.
Their own example, a refund asked for 45 days after purchase on a 30-day policy, answered all 3 questions in 277 ms with the same answers on every repeat:
- eligible: 2.2% yes
- route: refunds team, 86%
- urgency: Low, 63%
Longer states take about 0.6 s per decision at 1k tokens and 2.6 to 3.0 s at 4k.
These numbers are the floor. On Mac its linear-attention layers still run on PyTorch's generic kernels, so an optimized Metal path should make it faster. Their 33 ms figure is on a dedicated GPU.
If you're building agents, about 90 ms a decision is fast enough to check a policy before an agent acts or route tickets on your own machine, with no per-call bill. I didn't benchmark accuracy, but it gave the same answers on their refund example every run.
Introducing IOTA Decide 4B, our first public model.
Pareto optimal. Built different.
Trained across globally distributed GPUs with our SDK.
Open weights, Apache-2.0.
🤗 huggingface.co/macrocosm-os/…
const retweeted
The reason people reacted against centralized data centers is that they need to feel like they own a piece of the future
Decentralized AI and personal superintelligence is the way forward
A future where people own and control the extension to their cognition, both hardware and software, is needed
4* frontier model releases from Bittensor today.
ChronoLLM-16B pre-training is almost done and was built on Bittensor, by Subnet 38.
2 big updates:
1. ChronoLLM will release the largest Point-In-Time LLM out there, a 16B MoE.
2. Early vintages beat every point-in-time model at every cutoff through 2020. ChronoLLM should own the frontier for LLMs without lookahead.
ChronoLLM-16B is 8x bigger than its predecessor and scores 22% higher.
Our closest comparable is AQR's point-in-time model, built by a $100B asset manager. ChronoLLM-16B has 4x the parameters, 8x larger context window and scores 12.5% higher on the Open LLM benchmark (22% on the 2013 cutoff). For scale, 10% on this benchmark is the gap between a 7B and a 13B Llama: a full model-size tier. And it stays flat across every cutoff from 2013 to 2022. That’s what no lookahead looks like to us.
ChronoLLM-16B is also the first point-in-time model that can reason. Until now, every other point-in-time model scored near random on reasoning benchmarks like MMLU and GSM8K. ChronoLLM doesn't.
We’re here to stay.

























