🤖 Decentralised Training Miniseries, Day 1

Decentralised Training remains a topic I am incredibly bullish on. (I wrote an article about it here: https://www.antidoteblog.com/post/decentralized-training-the-1000x-opportunity-and-networks-competing-with-centralized-players

I want to do a daily mini series on what has been happening in the past few months and how you can potentially participate in the upside. Let us start with Nous Research.

🤖 Decentralised Training Miniseries, Day 1

Decentralised Training remains a topic I am incredibly bullish on. (I wrote an article about it here: https://www.antidoteblog.com/post/decentralized-training-the-1000x-opportunity-and-networks-competing-with-centralized-players

I want to do a daily mini series on what has been happening in the past few months and how you can potentially participate in the upside. Let us start with Nous Research.

https://nousresearch.com/

Latest on Nous: 
- End of April, it was reported that Nous raised their Series A from Paradigm https://cointelegraph.com/news/nous-research-raises-50m-paradigm-decentralized-ai-solana
- Nous also launched the testnet of Psyche, their decentralised training network based on Solana. 
- Along this launch, they started with the decentralised training run of consilience, a 40b parameter DeepSeek-style LLM.  https://psyche.network/runs/consilience-40b-1/0
-  When starting the pre-training run, Nous opened a $500k donation pot that was filled within 45 minutes. Nous is explicitly saying that these are donations, but there is a good possibility that this qualifies for a future Airdrop. The team mentioned that future donation pots will likely be pre-announced.

Thoughts on antidoteblog.com

Agents and Intents @NEAR | prev investments and BD 
@ZkSync, New Enterprise Associates and Balderton Capital

- The current run is being trained on 128H100s, split across 8xH100 clusters. Because of tech restrictions, the 8xH100 clusters need to be co-located. At the moment, the network is not permissionless, so you need to apply to participate. 
- This could be a potential opportunity for an investment DAO that buys and maintains a compute clusters in order participate in the pre-training. Let’s say a cluster costs $250k, and the $500k donations would get distributed among the cluster per run, this would yield 30k per cluster / run. At 4 training runs / year and 40% depreciation, you would be comfortable in the plus even minus energy cost. The potential upside with an airdrop of course could be much higher.

- How to potentially participate:  
- Put some capital into the next donation pot 
- If you have access to compute, participate in the training runs (or raise the capital to do so)
-  Use the Developer APIs of other Nous models (such as Hermes)