🤖 Decentralised Training Miniseries, Pt.2

Decentralised training is buzzing. In each part of this miniseries, I explore one of the major startups in this space, see what they have been up to recently and how you can potentially participate in the upside. Today we are looking at Prime Intellect.

🦋 Prime Intellect
The team has been incredibly productive and keeps churning out new stuff
- At the beginning of the year, Prime Intellect raised a $15M Series A round led by FoundersFund https://www.primeintellect.ai/blog/fundraise
- They have also finished a decentralised training run of Intellect-2, a reasoning-focused large language model (LLM) designed to excel in mathematics and coding tasks. It builds on the QwQ-32B model and improved reasoning capabilities compared to its base model.  https://app.primeintellect.ai/intelligence/intellect-2
- Intellect-2 worked by creating 300k reasoning traces via a parallelised, decentralised compute setup. The results then get used to adjust the weights of the model, improving its capabilities. The benefits:  Potential for heterogeneous compute, meaning contributors can use varied hardware.

New thing  |
Farcaster Pro User #235 |
Agents and Intents @NEAR | prev investments and BD 
@ZkSync, NEA and Balderton Capital

💶 How to participate in the upside: 
- The current Intellect-2 training run is closed. The good news: The team announced that more training runs will follow, such as Intellect-3 and Synthetic-2
- Even better news: The team mentioned that these training runs should be open to consumer-grade hardware. If you have a higher-end consumer Grade GPU, this could qualify you to participate. 
- Training runs often close fast in terms of compute participation. In oder to catch the new training runs, go to the Prime Intellect Discord, and take the ‘🛠️ Compute Provider’ role. This will notify you of future runs and let you apply in time. As always, the teams don’t promise any airdrops, but logically this maximises your chance to participate in a potential future airdrop. 
- In the meantime, you can also make use of their decentralised inference supply or use their Intellect-2 model in this chat.

If you are interested to learn more about decentralised training and why it is exciting, check out my blog article 
https://www.antidoteblog.com/post/decentralized-training-the-1000x-opportunity-and-networks-competing-with-centralized-players

💶 How to participate in the upside: 
- The current Intellect-2 training run is closed. The good news: The team announced that more training runs will follow, such as Intellect-3 and Synthetic-2
- Even better news: The team mentioned that these training runs should be open to consumer-grade hardware. If you have a higher-end consumer Grade GPU, this could qualify you to participate. 
- Training runs often close fast in terms of compute participation. In oder to catch the new training runs, go to the Prime Intellect Discord, and take the ‘🛠️ Compute Provider’ role. This will notify you of future runs and let you apply in time. As always, the teams don’t promise any airdrops, but logically this maximises your chance to participate in a potential future airdrop. 
- In the meantime, you can also make use of their decentralised inference supply or use their Intellect-2 model in this chat. https://chat.primeintellect.ai/chat/4ed774db-052b-40e2-ae88-d947a878686c