Volunteers Are Building a Free AI Model Without Big Tech
This could mean free, community-owned AI that anyone can use and trust.
Deep Dive
A project called CrowdGPT aims to create Crowd-v1, a collaboratively trained 1B-parameter transformer, without datacenters and for $0. According to the article, the project reached 85 registered contributors and led to 13B tokens trained in a week.
The goal is to show we don’t need huge Nvidia GPU clusters or millions of dollars to train a competitive Large Language Model, with the model size being community-driven. Anyone can submit data, the source code is auditable on GitHub, and the project is meant to be collaborative at every stage.
Key Points
- Volunteers are training a free AI model called Crowd-v1 using their own computers, not big data centers.
- In one week, 85 people trained 13 billion words, aiming for a 1 billion parameter model to rival existing ones.
- Anyone can contribute data or code, and the project is fully open-source and auditable.
Why It Matters
This could lead to free, community-owned AI that is transparent and not controlled by big corporations.