Bittensor: New Trends in Usage

Using Bittensor to outsource technological innovation and as an external incentivization layer

Image generated with Corcel, powered by Subnet 19

Bittensor is often described as a decentralized platform for launching digital commodity networks. While digital commodities are typically thought to only include compute, storage, and bandwidth, Bittensor extends well beyond these categories. A digital commodity network, in a broader sense, can refer to any network that offers a standardized digital task or service, governed by a clear and consistent incentive and validation framework.

This means that in addition to conventional subnets centered around web scraping, data storage, and cloud computing, Bittensor also supports subnets dedicated to specialized tasks and services such as creating AI models for specific modalities, fine-tuning open-source LLMs, and 3D content, image, and trade signal generation.

This ability to incentivize miners to provide a specific service using a standardized incentive and validation framework has led to teams getting creative in how they use Bittensor. Two emerging trends in what subnets are being used for are: 

1. Outsourcing technological innovation

2. As an incentivization layer for standalone networks

Outsourcing Technological Innovation

Recently, a trend has emerged where crypto teams are leveraging Bittensor to outsource the development of essential technologies that underpin their products or services. Rather than maintaining an in-house research and development team, these entities are turning to Bittensor. Both centralized and decentralized projects are creating competition markets as subnets, incentivizing contributors to work on specific problems that they define.

OpenKaito Subnet

Take Kaito, a centralized AI-powered search engine that caters to the crypto industry. Their goal is to make information within crypto more accessible by indexing crypto content and transforming unstructured data into a format that is both searchable and actionable.

Building a search engine presents numerous complexities, including data acquisition, indexing, ranking, and knowledge graph development. To address these challenges without maintaining an extensive in-house R&D department, the Kaito team launched the OpenKaito subnet on Bittensor. Here, the challenge of search relevance is framed as a miner-validator problem. Miners on the subnet submit ranked results for search queries, while validators apply a reward model to assess the quality of these miner responses. 

This approach enables Kaito to outsource essential R&D tasks, leveraging the collective expertise of contributors with domain-specific knowledge to build a decentralized search engine. Later down the line, Kaito aims to develop a search and analytics product on top of the subnet, which they intend to monetize.

MyShell and Virtual Subnet

MyShell and Virtual are two decentralized projects employing a similar strategy. MyShell focuses on the AI consumer layer, allowing users to craft personalized chatbots, including ones for famous characters or celebrities. To enhance their chatbots, the team plans to add voice functionalities. However, given the nascent stage of text-to-speech (TTS) technology and the absence of suitable solutions for custom voice models, MyShell launched a subnet to incentivize the development of open-source TTS models. This move allows them to shift their focus away from machine learning problems to other vital aspects of their network.

Virtual follows suit, but with a subnet focusing on incentivizing the development of audio-to-animation models.

Why Outsource to Bittensor?

Both MyShell and Virtual incentivize contributors through their protocols to contribute data and models for developing characters, custom chatbots, and to complete other tasks essential for their platforms' products and services. So why do they use Bittensor to drive the development of critical AI models underpinning their platform instead of doing so via their own protocols?

There are likely a couple of reasons:

1. Easier to Attract Contributors: It's challenging to attract experts with specific domain knowledge to contribute to an early-stage project, especially ML experts. However, Bittensor boasts a strong brand and an extensive network of miners/contributors with diverse expertise. Among these contributors are specialists in ML, who can seamlessly opt in to mining the subnets of projects like MyShell and Virtual and helping them with their objectives.

2. Immediate Value for Contributors: Contributors prefer to be immediately rewarded for their work in a valuable currency. TakeMyShell, for example, which doesn't have a live token. While they could provide points to contributors, serious contributors are unlikely to commit to substantial work based solely on the promise of future tokens without knowing their potential value. Even in cases where small projects do have tokens, leveraging Bittensor, where contributors earn TAO, a relatively more established token with a fair amount of liquidity, allows contributors to be immediately compensated in a stable manner.

Serving as a Network’s Incentivization Layer

One of the biggest challenges in launching a new network lies in scaling up the supply-side (the pool of miners contributing resources) to a critical mass before the demand-side (users) can begin utilizing the network's services. Crypto networks have proven effective in addressing this chicken-and-egg dilemma by incentivizing the supply-side with tokens for their presence and availability, even if they aren’t actively engaged in user tasks.

However, with the proliferation of AI and the surge in teams building AI resource networks and general digital commodity networks, attracting miners and bootstrapping the supply-side of a network has become increasingly challenging and competitive.

In this environment, Bittensor is uniquely positioned to become an external incentivization layer for networks, enabling a network to easily bootstrap its supply-side and focus solely on the execution layer of the protocol.

Case Study: Inference Labs

Inference Labs is working to bring AI on-chain through a Proof-of-Inference verification model that leverages zero-knowledge (zk) technology via an AVS on Eigenlayer. Importantly, they've also launched a subnet on Bittensor, Omron, specifically intended to bootstrap the supply of zk provers and model inferencers for their protocol. 

Essentially, Inference Labs is using Bittensor as the incentivization layer for their network’s supply-side in its initial stages.

The rationale behind leveraging Bittensor is straightforward: it is significantly easier to attract contributors to mine a subnet on an established network like Bittensor than it is to draw them into a new, standalone network. As mentioned above, Bittensor’s ability to provide immediate value to contributors is a major selling point. In addition, the network has thousands of miners who are already contributing to various subnets, making it seamless for them to opt in to new subnets due to their familiarity with the required resources and tasks for mining different digital commodity networks. 

Thus, launching a subnet on Bittensor allows Inference Labs to tap into this existing pool of skilled miners, accelerating the development and growth of their protocol. And accelerate it did. In just two weeks, assuming miners are operating with the minimum hardware requirements (which likely underestimates the actual capacity), the subnet collectively comprises 1900 CPU cores, 15 TB of RAM, and 90TB of storage, positioning the subnet as the largest zkML computing cluster.

In the future, Inference Labs plans to internalize the incentivization layer, wherein miners who directly contribute through their protocol will be rewarded with token incentives and network usage fees. However, even as Inference Labs transitions to its own incentivization mechanism, the subnet on Bittensor will persist, continuing to complement the protocol’s native supply-side indefinitely. In this capacity, Inference Labs’ network acts as an aggregator, sourcing zkML contributors from various channels, including the Bittensor subnet.

While some networks may opt to eventually internalize the incentivization layer, others may choose to permanently delegate this function to Bittensor, allowing them to focus exclusively on the execution layer.

How Does TAO Fit Into All this?

Within Bittensor, validators typically have exclusive access to the digital commodities produced by miners (more on this here). When a team launches a subnet to function either as an external incentivization layer or to outsource technological innovation for their network, the protocol or team has two options:

1. Become a Validator – This involves acquiring TAO and staking it to the specific subnet. The amount of bandwidth, or access to resources or services, a validator receives on a subnet is usually determined by the stake they hold. For instance, if a team operates a validator with 20% of the TAO stake of a subnet, it receives a corresponding 20% of the subnet’s bandwidth.

2. Pay Existing Subnet Validators for Bandwidth – Alternatively, teams have the option to pay existing subnet validators for their subnet bandwidth. This payment can be made in various currencies, such as fiat or stable coins, depending on the preferences of the validator. Taoshi is developing a Request Network that will seamlessly enable validators to monetize their bandwidth, allowing third parties to easily access the commodities of a subnet through APIs.

As the revenue potential of a subnet expands with increasing demand for its resource or service, validators begin competing to acquire TAO to secure additional bandwidth and prioritization from miners in order to earn a larger part of the revenue stream and enhance their operations. 

Ultimately, both options contribute to driving up demand for TAO within the Bittensor ecosystem. Given that TAO has a fixed token supply, this rising demand could consequently lead to an increase in its value.

As more talented miners join Bittensor’s mining network, more teams are drawn to launch a subnet to tap into the expert talent needed to solve their specific problems and source the digital resources they’re looking for. This creates a flywheel effect where talented miners competing over rewards attract higher quality subnet teams, which produce more valuable digital commodities, ultimately increasing demand and prompting validators to compete for TAO to secure bandwidth.

Final Thoughts

The prevailing theme among teams looking to launch a subnet on Bittensor seems to be the desire to tap into the mining community. One network, comprising contributors with expertise spanning machine learning, data science, trading, cloud computing, resource provisioning, and more, all seeking avenues to leverage their skills and resources in exchange for a common currency. At present, this appears to be Bittensor’s strongest moat.

What I find most interesting is how Bittensor can enable the development of initiatives that wouldn’t be feasible as standalone networks. For example, users can launch a subnet tailored to outsource specific components of their tech stack, much like deploying specialized microservices. For instance, a decentralized social network could offload the recommendation algorithm to Bittensor, similar to how projects are currently relying on Bittensor for the development and inferencing of AI models across different modalities.

As the subnet limit is raised in the near future, the prospect of centralized and decentralized projects hosting specific parts of their tech stack on Bittensor becomes increasingly plausible. I anticipate that this will lead to the emergence of a third category of use cases for Bittensor in the near future.

–

Thanks to Colin Gagich, Chris Abiaad, Anand Iyer, and Nick Hotz for their feedback on this piece.