# Distributed training reaches state of the art in 2-3 years

- **Speaker:** Greg Osuri
- **Attribution:** attributed
- **Said:** 2026-04-23
- **Theme:** Decentralized AI — https://www.gregosuri.com/predictions/decentralized-ai/
- **Source:** The Infrastructure Behind Agentic Finance, with Shashank Yadav and Greg Osuri of Akash Network (Fraction AI)
- **Watch at:** 00:26:49 — https://www.youtube.com/watch?v=59w3DyJWAQc&t=1609s
- **Canonical:** https://www.gregosuri.com/predictions/decentralized-ai/2026-04-23-distributed-training-reaches-state-of-the-art-in-2-6gku/
- **Cite as:** Greg Osuri, "Distributed training reaches state of the art in 2-3 years," The Infrastructure Behind Agentic Finance, with Shashank Yadav and Greg Osuri of Akash Network (Fraction AI), 23 Apr 2026, 00:26:49. https://www.youtube.com/watch?v=59w3DyJWAQc&t=1609s

## Quote — verbatim

> They trained a 72 billion parameter model which is better than Llama 70B... there are a lot of models that are trained in a fully distributed manner... that I believe will achieve SOTA in like two to three years. It's not there yet.

## Context — site annotation, not spoken

Citing frameworks from Pluralis, Gensyn, and Prime Intellect's OpenDiLoCo, plus a Bittensor subnet's 72B model, against the 3.2 Tbps NVLink bandwidth objection; he also mentions running Llama on his own home GPU. (Caption garbles "SOTA" as "sort.")