Bittensor (TAO) is the first AI coin listed on Binance this year. I thought it was the first step in the full opening of the AI track, but it turned out to be the "last step" in the short term. Since its launch on April 11, the price of TAO has plummeted and has not yet recovered.
Following the decline in the price of the coin, the community has been debating the validity of the project. It all started with a series of sharp questions raised by Eric Wall, co-founder of Taproot Wizards, on social media on March 30 about Bittensor (TAO), which has now been read nearly 2 million times.
Eric Wall's core point can be summarized as follows:
・Many miners in subnet 1 repeatedly execute the same language model to answer prompts, which is inefficient and wastes resources. One miner can complete the task, and there is no need for thousands of miners to work in parallel.
・The verification mechanism of Subnet 1 is too simple. It only compares the similarity of answers, which makes it easy for miners to cheat.
・Currently, Subnet 1 is only used internally and cannot be used by ordinary users, so it has no practical value.
・The Bittensor project is just hyping the concept of "decentralized AI" to deceive retail investors and inflate the token price.
Although these questions point directly to some of Bittensor's pain points, they are also biased and short-sighted. Multi-miner redundancy may seem inefficient, but it is actually the only way for distributed collaboration. Bittensor's goal is to build a global AI network, and redundancy is a necessary cost, not a design flaw.
The verification mechanism is still relatively rudimentary, but Bittensor has been actively improving it. The latest plan includes the introduction of the Commit-Reveal weight mechanism, which can effectively curb opportunistic plagiarism by delaying the disclosure of the weights submitted by miners.
Subnet 1 is the first subnet of Bittensor, and its positioning is mainly for internal training and testing. However, the Bittensor ecosystem has expanded to dozens of subnets for different application scenarios, which provide tangible value in search, medical, education, games and other fields. Simply classifying Bittensor as an "AI meme coin" and denying its value is itself an irrational and short-sighted approach.
Despite these doubts and challenges, Bittensor has not stood still. Instead, it continues to expand and improve its network. On May 12, Bittensor announced that it will add 4 subnet slots per week until it reaches the new 64 slot limit, with the goal of moving towards 1024 subnets this year.
As of now, Bittensor has 34 subnets, which cover multiple fields and fully demonstrate the potential and diversity of decentralized AI. Next, this article will introduce these subnets one by one from the six fields of content generation, data collection and processing, LLM ecosystem, decentralized infrastructure, DeFi and other applications, so that readers can have a comprehensive and clear understanding of the Bittensor ecosystem.
Content Generation
The subnet of the content generation category provides a platform for the generation and optimization of text, images, audio, and video.
Text Prompt (Subnet 1): Developed by the Opentensor Foundation, it is a decentralized subnet dedicated to text generation. It uses large language models (such as GPT-3, GPT-4, etc.) for prompts and reasoning, miners provide AI services, and validators are responsible for verifying the prediction results.
MyShell TTS (Subnet 3): Developed by MyShell, it focuses on text-to-speech (TTS) technology. This subnet develops and optimizes open source TTS models such as OpenVoice and MeloTTS. Miners are responsible for training models, and validators evaluate model performance, and are committed to creating high-quality open source TTS models.
Multi Modality (Subnet 4): Developed by Manifold, it focuses on multimodal AI systems that process and generate information across multiple data types and formats, including text, images, and audio.
Three Gen (Subnet 17): is a decentralized subnet focused on AI-driven 3D content generation. The Three Gen subnet uses AI technology to generate 3D models and content. Miners and validators are rewarded by contributing computing resources and verifying the quality of generated content, promoting the development of 3D content generation technology.
Cortex.t (Subnet 18): Developed by Corcel, it is a decentralized subnet focused on AI development and synthetic data generation.
Vision (Subnet 19): is a decentralized subnet focused on image generation and reasoning. The Vision subnet leverages the Distributed Scaled Inference Subnet (DSIS) framework to maximize the throughput of the Bittensor network, allowing miners to freely choose the technology stack to process requests and generate responses. Validators receive requests from the front end and distribute them to miners, evaluating their performance and making the image generation process more efficient.
Niche Image (Subnet 23): is a subnet focused on decentralized image generation. Niche Image supports multiple image generation models. Miners generate images by contributing computing resources and are rewarded based on quality. New models and features are constantly introduced to meet user needs.
TensorAlchemy (Subnet 26): is a subnet focused on human scoring and decentralized image generation. It evaluates the output of the image generation model through human scoring and rewards miners based on the score and the quality of the generated images. It plans to apply its technology in areas such as art creation and advertising.
Fractal (Subnet 29): Developed by Fractal Research, it is a decentralized subnet focused on text-to-video generation. This subnet uses a grid diffusion model and edge node inference technology to handle text-to-video tasks through distributed nodes.
WomboAl (Subnet 30): is a decentralized subnet focused on image generation and social sharing. The WomboAl subnet generates high-quality images through the Bittensor network and supports users to share images through applications such as WOMBO Dream and WOMBO Me.
Data collection and processing
Subnets in the data collection and processing category focus on decentralized data collection, storage, and analysis services. By building a distributed indexing layer and data processing framework, these subnets are able to process large-scale data sets and provide data support to other subnets and users.
Open Kaito (Subnet 5): Developed by Kaito AI, it aims to provide decentralized search and analysis services for Web3. This subnet builds a decentralized indexing layer to support intelligent search and analysis of Web3 content, and encourages miners to innovate and solve indexing tasks through Bittensor's incentive system.
Dataverse (Subnet 13): is a decentralized subnet focused on collecting and storing large amounts of data. The Dataverse subnet collects and stores data from various sources and provides data support to other subnets. Miners are rewarded with TAO tokens based on the amount of data they contribute, and validators regularly query and verify the correctness of the data.
Blockchain Insights (Subnet 15): is a decentralized subnet that focuses on converting raw blockchain data into structured graph models. This subnet provides data analysis query and result visualization functions, supports in-depth analysis of blockchain data, and users can perform customized queries.
Meta Search (Subnet 22): Developed by Datura-ai, it is a decentralized subnet focused on Twitter data analysis. Meta Search uses 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, it is a subnet focused on creating decentralized multimodal datasets, collecting video, audio, text and other data to support the research and development of general artificial intelligence (AGI). Miners are rewarded based on the data they contribute.
Conversation Genome Project (Subnet 33): Developed by Afterparty AI, it is a subnet focused on decentralized conversation data processing and personalized AI access. This subnet processes and indexes large amounts of conversation data in a decentralized manner, provides personalized AI access services, and miners are rewarded by contributing computing resources.
LLM Ecosystem
The subnetworks in the LLM Ecosystem category focus on training, fine-tuning, securing, and optimizing Large Language Models (LLMs).
Nous Finetuning (Subnet 6): Developed by Nous Research, it focuses on fine-tuning large language models (LLMs). This subnet rewards miners for fine-tuning LLMs using synthetic data, enables cross-subnet communication, and incentivizes miners by evaluating model performance.
Pretraining (Subnet 9): Developed by the Opentensor Foundation, it 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, it is a subnet focused on creating role-playing models. Dippy Roleplay incentivizes the community to create and optimize role-playing large language models (LLMs) in a decentralized manner. Miners and developers are rewarded with TAO tokens based on the quality and performance of the models they contribute.
LLM Defender (Subnet 14): Developed by Synapsec AI, it is a decentralized subnet focused on protecting large language models (LLMs) from various attacks. The LLM Defender subnet detects and prevents attacks on LLM applications through multiple analyzers and engines, leveraging the decentralized nature to provide a multi-layered defense mechanism.
NAS Chain (Subnet 31): is a decentralized subnet focused on Neural Architecture Search (NAS). NAS Chain uses genetic algorithms and distributed computing resources to optimize neural network architectures. Miners participate in NAS tasks by contributing computing resources and receive rewards based on their contributions.
Its AI (Subnet 32): is a decentralized subnet focused on detecting content generated by large language models (LLMs). This subnet uses the deberta-v3-large model to recognize text generated by LLMs, and is applied to multiple 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 improve the decentralization and stability of the network by providing distributed computing and storage resources.
Subvortex (Subnet 7): Encourage miners to run subtensor nodes through incentive mechanisms, enhancing the decentralization and stability of the Bittensor network. This subnet deploys nodes globally, has low latency and high redundancy, and lowers the threshold for participation.
Horde (Subnet 12): Developed by Backend Developers Ltd, it is a subnet focused on decentralized computing resource allocation. The Horde subnet distributes tasks to different miner nodes through distributed computing to improve the efficiency and speed of task processing. Miners are rewarded based on the computing resources they provide and the efficiency of task processing, and validators evaluate the quality of miners' work.
Filetao (Subnet 21): is a decentralized distributed storage subnet. FileTAO implements an efficient and secure storage system through a zero-knowledge proof space-time algorithm, supports multi-level verification mechanisms and cross-subnet communication, and miners are rewarded by contributing storage space.
Compute (Subnet 27): Developed by Neural Inτerneτ, it is a subnet focused on decentralized computing resource allocation. The Compute subnet provides a permissionless computing market, integrating multiple cloud platforms to form a unified decentralized high-level cloud computing infrastructure. Miners are rewarded with TAO tokens by contributing computing resources.
DeFi
Subnets in the DeFi category focus on the optimization and innovation of decentralized financial services, including liquidity staking, quantitative trading, yield optimization, and financial market prediction.
Omron (Subnet 2): Developed by Inference Labs, it aims to optimize and verify liquidity staking and re-staking strategies through artificial intelligence and machine learning technologies. Omron uses smart contracts and verification nodes to provide automated re-staking strategies and ensures the authenticity and security of the reasoning process through a zero-knowledge proof mechanism.
Proprietary Trading Network (Subnet 8): Developed by Taoshi, it focuses on decentralized quantitative trading signals. Miners contribute trading signals, covering multiple financial markets, and users can obtain high-quality trading signals.
Sturdy (Subnet 10): Developed by Sturdy Finance, it is a subnet focused on decentralized yield optimization. The Sturdy subnet allows miners to allocate assets to different strategy pools through smart contracts to achieve the highest yield. Miners are rewarded based on how much yield their allocation strategy generates, and validators evaluate miners' allocation strategies and score them based on yield performance.
Foundry SP 500 Oracle (Subnet 28): Developed by Foundry Digital LLC, it is a decentralized subnet focused on financial market prediction. The subnet incentivizes miners to predict the price of the SP 500 index and the prediction results are evaluated by validators.
other apps
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. The BitAds subnet distributes advertising tasks in a decentralized manner, and miners generate organic traffic by promoting advertising links and receive TAO token rewards.
BitAgent (Subnet 20): is a decentralized subnet focused on task and workflow management. BitAgent combines large language models (LLMs) with user-common applications to provide intelligent agent services to simplify daily tasks and workflow management. Miners compete based on performance and are rewarded with TAO tokens based on task completion.
Protein Folding (Subnet 25): Developed by the Opentensor Foundation, it is a decentralized subnet focused on protein folding research. Protein folding research is conducted through distributed computing resources, and miners are rewarded based on the computing power they contribute, providing a platform for biomedical research.
Healthi (Subnet 34): Developed by Healthi Labs, it is a decentralized subnet focused on using artificial intelligence (AI) to improve healthcare services. The Healthi subnet uses AI models for clinical prediction tasks and manages and processes medical data in a decentralized manner to ensure data security and privacy. Smart contracts simplify insurance processes and improve the efficiency of medical services.
Conclusion: Emerging Application Trends of Bittensor
As former Messari researcher Sami Kassab pointed out in a recent article, there are currently two emerging application trends for Bittensor: one is that projects outsource technological innovation to Bittensor subnets, such as Kaito AI outsourcing the development of search engines through Bittensor; the other is that projects use Bittensor as an incentive layer to quickly gather miner resources and provide digital goods for their networks, such as Inference Labs launching the Omron subnet to guide the supply of zk provers and model reasoners.
As Bittensor expands its subnet, more projects may choose to outsource specific components of the technology stack to Bittensor in the future, which will become the third largest application scenario for Bittensor. Bittensor is accelerating the professional division of labor in the AI industry and promoting the emergence of more original projects. With the increase in participants, the Bittensor ecosystem is expected to form a positive cycle and usher in a new stage of vigorous development.
