State of Bittensor #8

In recent months, Bittensor and $TAO have made remarkable strides. The launch of Bittensor's unique feature, the subnets, marked a significant milestone, and the market began to recognize the technological innovation that Bittensor represents.

The only blockchain for AI applications is rapidly becoming its universe. This has made it hard to keep up, so we have decided to write a monthly "State of Bittensor" to keep our community updated on the latest and greatest from Bittensor's growing open-source community of builders. Let's get to it.

TL/DR

• $TAO stats
• Subnets
• Subnet of the month
• New Bittensor-related tokens
• Closing thoughts

$TAO stats

$TAO price action

Over the past month, $TAO has ranged from $250 to $350. Based on the market cap, Bittensor is now the 47th biggest project.

$TAO and $WTAO holders

As the network grows, the number of $TAO (and $WTAO) accounts have grown. In total, there are now roughly 116,000 token holders, which is a 10% increase from August!

Staking stats

Currently, 79% of all circulating $TAO is staked. This is partially why $TAO is so volatile, as the floating supply is low, meaning there aren’t that many tokens available when demand increases. The staking APY is 16.7%, while validators rake in 18%.

Subnets

Bittensor subnets are specialized networks within the Bittensor ecosystem, each designed to focus on specific AI and machine learning tasks. These subnets create a decentralized environment where participants, such as miners and validators, contribute to and improve AI models.

Each subnet functions like a neural network, where miners provide computational resources and AI models, and validators assess the quality of these models based on performance. The system is incentivized through Bittensor’s native token, TAO, which rewards participants for their contributions. Subnets can specialize in various tasks, such as text generation, image creation, price prediction, and fine-tuning large language models.

The Bittensor team decided to raise the Subnet Limit from 36 to 45, aiming to enhance network stability and infrastructure. Previously, the 36 subnet limit led to frequent deregistrations, putting pressure on Subnet owners and miners to quickly turn profitable. This adjustment addresses those concerns, with plans to eventually reach 1024 subnets.

The decentralized nature of Bittensor subnets offers significant advantages, including enhanced security by removing single points of failure, efficient resource allocation for AI tasks, and open participation, which encourages innovation. This framework paves the way for a collaborative, global AI network that can evolve rapidly through contributions from a diverse range of participants.

Bittensor subnets are considered a promising development for AI due to their ability to democratize access to machine learning, allowing developers and researchers to contribute to a decentralized AI ecosystem while earning rewards for their efforts

Subnet in the spotlight: Masa

Masa has launched its AI Data Subnet on the Bittensor network. Masa, which focuses on creating Fair AI, allows contributors to earn by sharing data and has gained backing from Digital Currency Group, Binance, and others.

The Masa Bittensor Subnet aims to enhance AI applications by offering real-time and structured data from sources like social media, podcasts, and news outlets. This data helps improve Large Language Models (LLMs) by providing up-to-date information and context. The Masa Subnet supports AI developers with data for use cases such as personalized AI companions and trading signals.

Three main participants power the Masa Bittensor Subnet:

1. Workers, who contribute data and computational resources.
2. Validators, responsible for maintaining network integrity by verifying data and transactions.
3. Oracle Nodes, which allow developers to access AI data and services.

Masa introduces a dual-token reward system using both $MASA and $TAO, incentivizing participants. This makes it the first project in Bittensor to implement such a model. Validators and workers are rewarded based on the value of their contributions, ensuring high-quality inputs to the network.

Masa’s architecture is based on a robust mathematical framework. Participants stake tokens to participate, and their rewards depend on performance evaluations, which assess data quality, response times, and model outputs. Poor performance is discouraged through penalties or “slashing,” which leads to token losses. This framework promotes continuous improvement and high performance among contributors.

Paladin AI

Paladin AI integrates with the Bittensor network to enhance its AI capabilities, particularly in decentralized machine learning. Bittensor is a decentralized peer-to-peer network that incentivizes the creation and sharing of AI models through its token economy, centered around the TAO token. Paladin leverages Bittensor’s unique structure to optimize AI algorithm selection and improve data integration.

Paladin AI specializes in providing dynamic, real-time data from various sources such as Ethereum transactions, Messari API, and social media platforms like Twitter and TikTok. By connecting to Bittensor, Paladin ensures that it accesses an ever-evolving pool of AI models, benefiting from Bittensor’s decentralized and permissionless approach to AI development. This allows Paladin to continuously update its data sets and recalibrate AI outcomes, ensuring accurate and real-time insights, which are crucial for areas like financial predictions, smart contract auditing, and blockchain security.

The integration with Bittensor also allows Paladin to tap into the network’s incentive structure, where validators and miners contribute to the development and evaluation of AI models. Paladin benefits from Bittensor’s consensus mechanisms, which reward high-quality AI contributions, further pushing the boundaries of what decentralized AI can achieve​

Omega Labs

OMEGA Labs has introduced a multimodal AI system on Bittensor's decentralized network. Their Subnet 21, in collaboration with Subnet 24, focuses on creating Any-to-Any models that handle multiple data types, including images, audio, and video. This approach aims to accelerate advancements in Artificial General Intelligence (AGI) by leveraging diverse datasets and decentralized AI infrastructure.

The OMEGA multimodal models outperform other state-of-the-art models like Qwen-VL (7B) on key benchmarks for image understanding, such as RealWorldQA and MathVista. These models use ImageBind embeddings, which unify visual, audio, and text data into a latent representation, allowing AI to process information across different modalities. The integration with Bittensor’s Subnet 24, known for collecting large-scale, high-quality data, empowers OMEGA models to continuously learn from diverse and constantly updated datasets.

OMEGA Labs takes full advantage of Bittensor’s incentive mechanisms by rewarding miners for contributing valuable data, allowing researchers to train powerful, open-source AGI models. This decentralized framework promotes collaboration, encouraging AI experts to develop models that serve real-world applications like intelligent digital agents, video analysis, and immersive gaming experiences​

Closing thoughts

The Bittensor ecosystem is just getting started, with subnets gaining more and more traction. All 45 subnet slots have been filled in this short time, with the subnet ecosystem updates coming and expansion to 1028 subnets in the coming years.

Bittensor is rapidly becoming a microcosm, and we can’t wait to see what AI products get released next. It will also be interesting to see how long it will take before the subnet competition starts to become fierce, with subnets feeling the pressure of being replaced. We’ll keep you in the loop until next month!

If you’d like to receive next month’s State of Bittensor, subscribe to our newsletter here!