Besides mapping out the @opentensor eco, we want to point out the necessity to research into subnets.
It’s shifting from $TAO appreciation drives subnet growth towards subnet progress reinforces the value of the network and, in turn, supports $TAO itself.
On the @theallinpod this week, @chamath asked @nvidia CEO Jensen Huang about decentralized AI training, calling our Covenant-72B run "a pretty crazy technical accomplishment."
One correction: it's 72 billion parameters, not four. Trained permissionlessly across 70+ contributors on commodity internet. The largest model ever pre-trained on fully decentralized infrastructure.
Jensen's answer is worth hearing too.
One of the sharpest independent analyses of Covenant-72B we've seen. The author identifies what they call a "cognitive arbitrage window": crypto investors see "another open-source model" and shrug, while AI researchers who understand the benchmarks don't follow crypto. That gap produced a 2-day price lag after the announcement before the market caught up.
The piece is thorough on the technical details (SparseLoCo compression, the honest benchmark gap to Qwen2.5/LLaMA-3.1, why the trajectory matters more than any single number) and places Covenant-72B in the full history of decentralized training from GPT-JT to INTELLECT-1. Google Translate handles it well if you don't read Chinese.