GM TO EVERYONE ☀️

When Jensen Huang mentioned decentralized AI training, attention immediately shifted toward Bittensor.
But this had already been explored well before that moment.
Back in June 2025, @0G_labs published the DiLoCoX paper on arXiv, showing that large-scale model training across decentralized nodes was way more efficient.
They demonstrated 100B+ parameter training over standard hardware and typical internet, while improving communication efficiency by 357x compared to traditional methods.
There's also a key difference that often gets missed. Bittensor focuses on a specific trained network, while DiLoCoX is designed as a framework that can be used to train any model.
It's also part of a broader stack which is combining compute, storage, data availability, and chain.
Next stop: EthCC Cannes 2025 on April 1 📍