This morning, Yzi Labs announced its investment in Tensorplex Labs, and its Bittensor subnet project Dojo rose 500%. In the same month last year, the decentralized AI startup Tensorplex Labs completed a $3 million seed round of financing. A year later, Tensorplex Labs has brought contributions to the Bittensor ecosystem including Dex, data dashboards, AI, etc. This article will briefly analyze what Tensorplex Labs, a Singaporean company that was established only two years ago, has done to become a core participant in the Bittensor ecosystem.
Bittensor’s DeFi Core
Pledge loan
Before the release of dTao, the ecosystem of Bittensor was actually very scarce. Tensorplex contributed core DeFi products in the TAO ecosystem, one of which was Tensorplex Stake. As a cross-chain staking product, it launched stTAO, an LSD ERC 20 Liquid Staking Derivative token, which allows EVM investors to participate in the TAO ecosystem and enjoy staking rewards without downloading a Substrate-compatible wallet, and manually entrust. At the same time, the pledged assets can be bridged to networks such as Bittensor to participate in the operation of verification nodes. For this reason, Tensorplex has also established a bridge between Substrate and EVM to minimize trust. In addition, lending protocols such as TaoBank have been incubated, allowing users to obtain interest-free loans with pledged TAO as collateral. The peak TVL of the pledge reached more than 30 million US dollars, which also made Tensorplex one of the largest verification nodes on Tao.
Exchange + Data Dashboard Integration Backprop
Just last month, dTAO was officially launched and @BackpropFinance was also officially launched. At this point in time, as one of the three Dex that can trade dTAO, hotkeyswap, taostats, in addition to allowing users to trade various Alpha subnets, it also provides a fairly comprehensive data dashboard, integrating various types of products such as trading dashboards, address analysis, on-chain information query, and market analysis into the same product. The familiar UI and the perfect data dashboard have greatly reduced the threshold for new users to join the TAO ecosystem, which also makes Backprop the most popular Dex product of Bittensor.
Development of the AI ecosystem
LLM
Tensorplex has made many attempts in large models and created a similar model that surpassed all open source in the decentralized LLM pre-training competition of @MacrocosmosAI Subnet 19. Running validators on more than 60 Bittensor subnets and AI chains such as @flock_io and @AlloraNetwork , the latest top model Sumo was released on Bittensor Subnet 9 since 2023, and after increasing the parameter limit to 7 billion in April 2024, Sumo-T 9-7 B even outperformed well-known models such as Falcon 7 B and Llama 2 7 B.
Tensorplex Stream
Tensorplex has also tried agents. Tensorplex and @Corcel_X Subnet 18 built the first Web3 RAG chatbot Tensorplex Stream, which obtains information from cryptocurrency podcast records to train AI agents and releases an agent that aggregates and summarizes the latest cryptocurrency information. In addition to Chatbot, Podcasts have also been developed to allow AI to speak the information it has collected. Using the advantages of RAG to collect the latest information in the rapidly developing cryptocurrency market has also been widely used.
Dojo Subnet
Dojo Subnet has become the most shining subnet in dTAO in the past 24 hours, and it has soared 500% after YziLabs announced its investment in Tensorplex Labs. Dojo is a data collection platform, but it is not only a data collection platform, but also a comprehensive ecosystem that joins humans as observational third parties. Human intelligence can directly affect AI development, training and agent operations on the chain and in the real world.
The main workflow is that anyone can use LLM to generate unique synthetic tasks (which may be game physics feedback, quality of generated images, physics algorithms, etc.). Collect artificial feedback data through Dojo and add response obfuscation technology.
The miners will then distribute the tasks, and the job of the human verifier is to confirm some of the basic facts generated by the LLM, by determining the quality of the responses provided by each contributor and distributing rewards based on multi-dimensional indicators: manual consistency weighted Cohens Kappa, quantitative stability intraclass correlation coefficient ICC, factual fidelity Spearman correlation, these data will be used to improve and optimize the open source model.
Dojo will use technical means to prevent sybil attacks and ensure that contributors are real humans. When the system matures, it will also add the role of preventing attackers and form a contribution chain. So far, Dojo has collected more than 3 million human-generated data points.
Tensorplex Labs investment by Yzi Labs is just a microcosm of the Bittensor ecosystem being gradually accepted by the market. The halal Bittensor usually does not receive much attention in the market. In fact, as an old AI ecosystem, each project party in the ecosystem has a considerable degree of ability and patience, giving people a serious building atmosphere. Although there are still problems such as shells or dTAO systems in the subnet, we still expect Bittensor to output more excellent products, and BlockBeats will continue to pay attention.
