The best AI teams should build on Bittensor
Centralized AI is the biggest threat to humanity.
Corporations like OpenAI, Microsoft, Google are racing to build a superintelligent AI. They are incentivized by capitalism to make trillions for their shareholders and keep model weights behind closed doors.
They control not just the underlying model, but also the distribution channels to humanity. Google is the world’s most visited website and ChatGPT is the world’s most popular AI product.
Do we want the world’s most potent technology to be governed by a handful of greedy corporations? Is there an alternative movement that world-class AI talent should support?
Introducing Bittensor
Bittensor facilitates the production of various open-source AI commodities (for eg. AI models or datasets). There are 3 ways for an AI engineer to participate in Bittensor. They can be either a:
- Subnet Owner: owners define the tasks and rules that miners and validators must follow.
- Miner: miners are AI engineers that perform tasks defined by the subnet owners. They contribute compute and intelligence to solve problems while open-sourcing their solutions.
- Validator: validators quality check the miners work.
For example, the “Pre-Training Subnet 9” rewards miners for producing pretrained Foundation-Models on the Falcon Refined Web dataset. Tensorplex Labs is one of hundreds of miners on the subnet and they recently managed to open-source a SOTA model on top of Subnet 9, beating out notable 7B pre-trained models like Meta’s LLama 2 on various benchmarks.
Tensorplex is just one of several ultra-talented AI teams building on Bittensor. Others include:
- MyShell: a team of MIT/Oxford PhD’s that have open-sourced SOTA models across multiple modalities (this one has 26K+ GitHub stars 🤯),
- Kaito AI: a Sequoia-backed team of AI researchers
- Omega Labs: have built the world's largest multi-modal dataset on top of subnet 24 and have a great write-up on their long-term goals with the protocol
- WOMBO: a consumer AI studio with 150M+ downloads
This is a non-exhaustive list and each day 100s of world-class engineer join the Bittensor Discord to contribute to the open-soruce AI movement. There's also epic AI products like Corcel, which are built on top of a single subnet.
Sounds cool… But epic open-source software like Linux was built without Bittensor… Why do we need this for open-source AI?
AI is different from Web 2.0 software like Linux. Creating Linux requires smart people to dedicate just their time. AI is different - it requires incredibly smart people to dedicate their time, compute and data. Compute and data are expensive, discouraging AI teams from open-sourcing their work.
Bittensor’s incentive mechanism rewards contributors with Tao cryptocurrency for providing their time, compute and data. This approach fosters collaborative open-source AI development, while ensuring contributors are fairly compensated for their efforts.
But wait! What's in it for me? Open-sourcing AI is cool, but I don’t have time for this.
Currently, the financial rewards to build on Bittensor are incredible. Every year, millions of dollars are distributed amongst miners, validators and subnet owners. Tao (the token Bittensor is built on) was recently listed on Binance, making the rewards liquid.
This sounds too good to be true… where is all this money coming from?
Bittensor’s current subnet system started in October 2023. To attract world-class teams to the protocol 7,200 TAO tokens are minted daily and then distributed between miners, validators and subnet owners based on their performance.
This means that builders on Bittensor today are rewarded through inflation. This inflation is unsustainable over the long-term, but in the short-term is attracting real talent to the network. In many ways, protocol inflation is similar to venture capital inflation. My hope is that this inflation attracts the biggest AI companies of tomorrow to build on Bittensor today, open-sourcing SOTA AI commodities along the way.
Bittensor is in its infancy. It is an ambitious experiment and we will not know if it has succeeded or failed for many years. To ensure its success we need to attract the world’s most talented AI developers as soon as possible.
How do I get involved?
A great place to start would be the Bittensor Dev Docs and the Bittensor Discord Server. Each channel in the discord server is operated by a subnet owner and in the “pinned” section of each subnet channel you will find detailed info on how to get started as a miner.
TaoStats is a great website to visualize the whole ecosystem. There’s also incubators for teams looking to launch a new subnet - 0xAI is just one of many!
Bittensor offers a great opportunity to contribute to the open-source AI movement and my hope with this blog is to demystify the protocol for many!
My next few posts will outline how massive cashflow generating AI businesses will eventually exist on the protocol. And I will also have some posts outlining how to build a subnet from scratch, from my experience working on Dippy Subnet 11.