If the threat of financial centralization is what brought you to crypto, AI centralization should have you shitting bricks 🧱 Decentralized AI training is coming to Solana thanks to @NousResearch and their new network, Psyche. I dive into the details in our latest report 👇

The spending on AI from hyperscalers like OpenAI and Anthropic continues to go parabolic. And without some clever tricks, the headwinds faced by a hodge-podge collection of globally-distributed GPUs can seem overwhelming.

One major struggle is bandwidth: Data-centers have GPUs directly connected, communicating up to 1000x faster than internet speeds. Techniques like DisTrO from Nous even the odds by minimizing the chatter between GPUs.

You can even start to stack on Reinforcement Learning, which DeepSeek showed to be a huge compute-saver. The inherent asynchronous nature of RL also conveniently lends itself to distributed training.

And (for once, unironically) blockchain makes this better. ✅Permissionlessness: allowing anyone to contribute compute ✅Resilience and uptime: no reliance on centralized infrastructure ✅Token incentives: coordinate and reward contributors for improving the network

There’s so much more to go through, so go read the full report by following the link below (creating an account is free btw) and let us know what you think!

Link: mealdeal.blocmates.com/researc...