How we think about Bittensor
The recent criticism we’ve seen on X made us question what we thought about Bittensor. Instead of discouraging us, it made us shift our views on some aspects, and reinforce our conviction on others. We wanted to share these thoughts.
1. Most people still can’t see the forest for the trees
When talking about Bittensor, some people still look at how one subnet works and draw conclusions about the whole network.
All subnets are different from one another, and the way they work has nothing to do with how the main Bittensor network does.
It’s also important to note that subnets compete with each other for a spot in Bittensor. The less interesting ones regularly get deregistered to make place for more promising ones. This is actually the role of the Bittensor network: ensuring a healthy and beneficial competition between subnets.
The range of possibilities when designing a subnet is so wide that they can go from very shitty to incredibly promising. As the subnets have been introduced recently, it’s not a surprise we are still seeing an important turnover today.
2. Calling Bittensor an AI project is misleading
One view we have that doesn’t seem to be shared by most people in or outside Bittensor, is that Bittensor isn’t an AI project per se. The Bittensor network’s goal is to allocate incentives efficiently between the subnets.
A subnet is not a specific AI model either. It is a network in which miners need to find the best solution to a specific problem. Validators rate them, and they receive rewards according to those ratings.
That a specific subnet’s task is better solved using an AI model or not has nothing to do with Bittensor. This is typically determined when the subnet task and reward model are defined.
For example, the goal of subnet 27 is to “enable higher-level cloud platforms to offer seamless compute composability across different underlying platforms”. The task of miners is to make sure that computation is accessible in a seamless and efficient way, and spread across different cloud providers. There would be no point in using an AI model in that case.
In subnet 8, called Proprietary Trading Network, miners have to express trading signals (long-flat-short) and rewarded according to their performance. In that case, using an AI model could definitely be helpful. However, a miner could also choose a much simpler strategy that might also perform well.
In other cases, the goal of the subnet is directly related to AI. In NOUS-Research’s subnet 6, miners are tasked to fine-tune large LLMs with custom, open-source models.
Though AI can be well suited to solve some subnets’ tasks, or is a key component of others, we believe labelling Bittensor as an AI project is:
- Misleading for people discovering the project.
- Not reflecting its real nature and potential.
The fact that any tool can potentially be used to work on a subnet task, from an AI model to a simple human input, is exactly what makes Bittensor so powerful.
3. TAO tokenomics is underrated
Bittensor more or less follows the issuance schedule of bitcoin, with TAO being limited to 21 million units. Since inception, one TAO is issued at each block every 12 seconds, making it 50 every 10 minutes, and halvings are set to reduce this rate. Some saw it as a tribute to Satoshi’s invention, others as a cringe attempt to be recognized.
When looking behind the curtains, there is actually one key difference. Instead of letting the block numbers determine halvings, they happen once half the supply of what is left to be minted since the last halving has been issued. To make it clearer:
And so on.
What makes this particularly interesting, is that TAO has to be burnt by those that want to register as a miner or validator in a subnet. This reduces the circulating supply of TAO, delaying the next halving.
In other words, as more people want to participate in Bittensor as miners or validators, the next halving, and its effect on price, is delayed. This creates a situation where the transition to a hard asset is gradual, and leaves more time for adoption to take place.
We are a team of three looking for funding to build on Bittensor. If you are an individual or a VC that might be interested, you can reach out at mentatminds@protonmail.com.
References :
