Bittensor (TAO) became the first AI coin listed on Binance this year. It was expected to be the first step in the comprehensive development of the AI sector, but it turned out to be a short-lived endeavor. Since its launch on April 11th, the price of TAO has been on a downward trend and shows no signs of recovery.
Alongside the price decline, there has been an increasingly intense debate within the community about the project’s effectiveness. It all started on March 30th when Eric Wall, co-founder of Taproot Wizards, raised a series of sharp questions about Bittensor (TAO) on social media. These questions have now reached nearly 2 million reads.
Eric Wall’s main points can be summarized as follows:
– In Subnet 1, many miners repeatedly execute the same language model to answer prompts, which is inefficient and wasteful of resources. It is unnecessary to have thousands of miners working in parallel when one miner can complete the task.
– The validation mechanism in Subnet 1 is too simple, only comparing the similarity of answers, making it easy for miners to take advantage and cheat.
– Currently, Subnet 1 is only running internally and is not accessible to ordinary users, thus lacking practical value.
– The Bittensor project is simply a hype around the concept of “decentralized AI,” deceiving retail investors and artificially inflating the token price.
While these criticisms directly address some of Bittensor’s weaknesses, they may also be biased and overlook the bigger picture. The redundancy of multiple miners may seem inefficient, but it is a necessary cost of distributed collaboration. Bittensor aims to create a globally scaled AI network, and redundancy is a necessary cost, not a design flaw.
The validation mechanism is still relatively basic, but Bittensor is actively working on improvements. The latest plans include the introduction of the Commit-Reveal weighting mechanism, which can effectively curb cheating and plagiarism by delaying the disclosure of weights submitted by miners.
Subnet 1, as the first subnet of Bittensor, is primarily focused on internal training and testing. However, the Bittensor ecosystem has expanded to dozens of subnets targeting different application scenarios, providing real value in areas such as search, healthcare, education, and gaming. Categorizing Bittensor simply as an “AI meme coin” and denying its value is an irrational and shortsighted approach.
Despite these criticisms and challenges, Bittensor has not stood still. On May 12th, Bittensor announced that it will add four subnet slots per week until it reaches the new limit of 64 slots, with the goal of reaching 1024 subnets this year.
Currently, Bittensor has 34 subnets covering various fields, showcasing the potential and diversity of decentralized AI. In the following sections, we will introduce these subnets one by one in six areas: content generation, data collection and processing, LLM ecosystem, decentralized infrastructure, DeFi, and other applications. The aim is to provide readers with a comprehensive understanding of the Bittensor ecosystem.
Content Generation:
Content generation subnets provide platforms for generating and optimizing text, image, audio, and video.
Text Prompt (Subnet 1): Developed by Opentensor Foundation, this decentralized subnet is specifically designed for text generation. It utilizes large language models like GPT-3 and GPT-4 for prompts and inference. Miners provide AI services, while validators are responsible for verifying the prediction results.
MyShell TTS (Subnet 3): Developed by MyShell, this subnet focuses on text-to-speech (TTS) technology. It develops and optimizes open-source TTS models like OpenVoice and MeloTTS, with miners responsible for training the models and validators evaluating their performance to create high-quality open-source TTS models.
Multi Modality (Subnet 4): Developed by Manifold, this subnet specializes in multi-modal AI systems that process and generate information across various data types and formats, including text, images, and audio.
Three Gen (Subnet 17): This decentralized subnet focuses on AI-driven 3D content generation. Miners and validators contribute computing resources and verify the quality of generated content to receive rewards, driving the development of 3D content generation technology.
Cortex.t (Subnet 18): Developed by Corcel, this subnet focuses on AI development and synthetic data generation.
Vision (Subnet 19): This decentralized subnet focuses on image generation and inference. It utilizes the Distributed Scale Inference Subnet (DSIS) framework to maximize the output capacity of the Bittensor network. Miners are free to choose their preferred technology stack to handle demands and generate responses. Validators receive requests from the frontend and distribute them to miners, evaluating their performance to make the image generation process more efficient.
Niche Image (Subnet 23): This subnet specializes in decentralized image generation. It supports various image generation models, with miners rewarded based on the quality of the generated images, continuously introducing new models and features to meet user demands.
TensorAlchemy (Subnet 26): This subnet focuses on human rating and decentralized image generation. It evaluates the output of image generation models through human ratings and rewards miners based on the ratings and the quality of the generated images, with plans to apply its technology in fields like art creation and advertising.
Fractal (Subnet 29): Developed by Fractal Research, this decentralized subnet focuses on text-to-video content generation. It uses grid diffusion models and edge node inference technology to process text-to-video tasks across distributed nodes.
WomboAl (Subnet 30): This subnet specializes in image generation and social sharing. It generates high-quality images through the Bittensor network and supports applications like WOMBO Dream and WOMBO Me for users to share images.
Data Collection and Processing:
Subnets in the data collection and processing category focus on decentralized data collection, storage, and analysis services. By building distributed indexing layers and data processing frameworks, these subnets can handle large-scale datasets and provide data support to other subnets and users.
Open Kaito (Subnet 5): Developed by Kaito AI, this subnet aims to provide decentralized search and analysis services for Web3. It constructs a decentralized indexing layer to support intelligent search and analysis of Web3 content, encouraging miners to innovate and solve indexing tasks through the Bittensor incentive system.
Dataverse (Subnet 13): This subnet focuses on collecting and storing large amounts of data in a decentralized manner. It collects and stores data from various sources and provides data support to other subnets. Miners receive TAO token rewards based on the amount of data they contribute, while validators regularly query and verify the accuracy of the data.
Blockchain Insights (Subnet 15): This decentralized subnet specializes in transforming raw blockchain data into structured graph models. It provides data analysis queries and result visualization functions, supporting in-depth analysis of blockchain data, allowing users to perform customized queries.
Meta Search (Subnet 22): Developed by Datura-ai, this subnet focuses on Twitter data analysis. Meta Search utilizes AI technology to conduct in-depth analysis of Twitter data, providing real-time data access and sentiment analysis to help users understand public sentiment and make data-driven decisions.
Omega Labs (Subnet 24): Developed by Omega Labs, this subnet focuses on creating decentralized multimodal datasets, collecting data such as videos, audios, and texts to support general artificial intelligence (AGI) research.Development support is provided, and miners are rewarded based on their contributions.
The Conversation Genome Project (Subnet 33), developed by Afterparty AI, is a subnet that focuses on decentralized conversation data processing and personalized AI access. This subnet processes and indexes a large amount of conversation data in a decentralized manner, providing personalized AI access services. Miners are rewarded for contributing computing resources.
LLM Ecosystem
Subnets in the LLM ecosystem category focus on the training, fine-tuning, protection, and optimization of large language models (LLMs).
Nous Finetuning (Subnet 6), developed by Nous Research, focuses on fine-tuning LLMs. Miners are rewarded for fine-tuning LLMs using synthetic data, enabling cross-subnet communication, and incentivizing miners through model performance evaluation.
Pretraining (Subnet 9), developed by Opentensor Foundation, focuses on pretraining large language models. Miners train models on the Falcon Refined Web dataset and improve model performance through continuous benchmarking and validation mechanisms.
Dippy Roleplay (Subnet 11), developed by Impel, is a subnet that focuses on creating role-playing models. Dippy Roleplay incentivizes the community to create and optimize role-playing LLMs in a decentralized manner. Miners and developers are rewarded with TAO tokens based on the quality and performance of their contributions.
LLM Defender (Subnet 14), developed by Synapsec AI, is a decentralized subnet that focuses on protecting large language models (LLMs) from various attacks. LLM Defender subnet detects and prevents attacks on LLM applications using multiple analyzers and engines, providing multi-level defense mechanisms through its decentralized features.
NAS Chain (Subnet 31) is a decentralized subnet that focuses on Neural Architecture Search (NAS). NAS Chain optimizes neural network architectures using genetic algorithms and distributed computing resources. Miners contribute computing resources to participate in NAS tasks and are rewarded based on their contributions.
Its AI (Subnet 32) is a decentralized subnet that focuses on detecting content generated by large language models (LLMs). This subnet uses the deberta-v3-large model to identify text generated by LLMs, and is applied in various scenarios such as machine learning, education, and social media. Validators use The Pile dataset to ensure the accuracy and reliability of the detection system.
Decentralized Infrastructure
Subnets in the decentralized infrastructure category enhance the decentralization and stability of the network by providing distributed computing and storage resources.
Subvortex (Subnet 7) incentivizes miners to run subtensor nodes, enhancing the decentralization and stability of the Bittensor network. This subnet deploys nodes globally, with low latency and high redundancy, reducing barriers to participation.
Horde (Subnet 12), developed by Backend Developers Ltd, focuses on decentralized computing resource allocation. Horde subnet assigns tasks to different miner nodes using distributed computing to improve task processing efficiency and speed. Miners are rewarded based on their provided computing resources and task processing efficiency, and validators assess the quality of their work.
Filetao (Subnet 21) is a decentralized distributed storage subnet. FileTAO implements an efficient and secure storage system using zero-knowledge proof space-time algorithms, supporting multi-level verification mechanisms and cross-subnet communication. Miners are rewarded for contributing storage space.
Compute (Subnet 27), developed by Neural Inτerneτ, focuses on decentralized computing resource allocation. Compute subnet provides a permissionless computing market, integrating multiple cloud platforms into a unified decentralized high-level cloud computing infrastructure. Miners are rewarded with TAO tokens for contributing computing resources.
DeFi
Subnets in the DeFi category focus on optimizing and innovating decentralized financial services, including liquidity staking, quantitative trading, yield optimization, and financial market prediction.
Omron (Subnet 2), developed by Inference Labs, aims to optimize and validate liquidity staking and restaking strategies using artificial intelligence and machine learning technologies. Omron provides automated restaking strategies through smart contracts and validation nodes, ensuring the authenticity and security of the inference process through zero-knowledge proof mechanisms.
Proprietary Trading Network (Subnet 8), developed by Taoshi, focuses on decentralized quantitative trading signals. Miners contribute trading signals covering multiple financial markets, and users can access high-quality trading signals.
Sturdy (Subnet 10), developed by Sturdy Finance, is a subnet that focuses on decentralized yield optimization. Sturdy subnet allows miners to allocate assets to different strategy pools through smart contracts to achieve the highest yield. Miners are rewarded based on the generated yield from their allocation strategies, and validators evaluate their allocation strategies and score them based on performance.
Foundry SP 500 Oracle (Subnet 28), developed by Foundry Digital LLC, is a decentralized subnet that focuses on financial market prediction. This subnet incentivizes miners to predict the price of the SP 500 index, and their predictions are evaluated by validators.
Other Applications
Subnets in the other applications category cover areas such as ad distribution, task management, protein folding research, and healthcare.
BitAds (Subnet 16) is a decentralized and incentivized advertising subnet. BitAds subnet distributes advertising tasks in a decentralized manner, and miners generate organic traffic by promoting advertising links to earn TAO token rewards.
BitAgent (Subnet 20) is a decentralized subnet that focuses on task and workflow management. BitAgent combines large language models (LLMs) with commonly used applications to provide intelligent agent services, simplifying daily tasks and workflow management. Miners compete based on performance and earn TAO token rewards based on task completion.
Protein Folding (Subnet 25), developed by Opentensor Foundation, is a decentralized subnet that focuses on protein folding research. Miners are rewarded based on their contributed computing power, providing a platform for biomedical research.
Healthi (Subnet 34), developed by Healthi Labs, is a decentralized subnet that focuses on using artificial intelligence (AI) to enhance healthcare services. Healthi subnet utilizes AI models for clinical prediction tasks and manages and processes healthcare data in a decentralized manner to ensure data security and privacy. Smart contracts simplify the insurance process and improve the efficiency of healthcare services.
Conclusion: Emerging Trends in Bittensor’s Applications
As noted by former Messari researcher Sami Kassab in a recent article, Bittensor is currently experiencing two emerging trends in its applications. Firstly, projects are outsourcing technical innovations to Bittensor subnets, such as Kaito AI outsourcing search engine development to Bittensor. Secondly, projects are using Bittensor as an incentive layer to quickly gather miner resources and provide digital goods supply to their networks, such as Inference Labs launching the Omron subnet to guide the supply of zk provers and model inferencers.
As Bittensor expands its subnets, there may be more projects choosing to outsource specific components of their technical stack to Bittensor, creating a third application scenario for Bittensor. Bittensor is accelerating the specialization of the AI industry and driving the emergence of more original projects. With an increasing number of participants, the Bittensor ecosystem is expected to form a positive feedback loop, ushering in a new stage of vibrant development.
