ConnitoAI Bittensor Subnet 102 dashboard

ConnitoAI is presenting Bittensor Subnet 102 as a decentralized training network for large language models, with a user dashboard expected around May 26, according to a May 22 social-media thread and the project’s own website. Connito says it is building training infrastructure for “100B+ Parameter Models,” while Bittensor’s subnet page describes the network as a distributed Mixture-of-Experts, or MoE, system. (connito.ai) The pitch is straightforward: instead of training one large model on centralized hardware, Connito splits work across specialized “experts” run by separate participants on the Bittensor network. (connito.ai) The company’s GitHub repository says Subnet 102 distributes expert groups across independent miners, while validators coordinate updates, aggregate contributions and score miners for TAO-denominated rewards. The dashboard mention matters because it points to a consumer-facing layer on top of what has so far been described mainly as subnet infrastructure. (connito.ai) The May 22 thread said early beta access would accompany the dashboard launch, though that timing was not independently confirmed on Connito’s website in the material reviewed. ### What is ConnitoAI actually saying it built? Connito’s website says the company is training “100B+ Parameter Models” and doing so “Cheaper and Better” through expert decentralization. (github.com) The site says contributors train specialized expert modules that are then aggregated into larger systems, rather than relying on “massive centralized compute.” A separate Connito explainer says the platform is designed to turn customer data into “production-ready, domain-specific AI.” That page says the system trains isolated expert modules that can be slotted into a shared architecture, and that each new engagement adds experts to a growing shared library. (connito.ai) ### How does Subnet 102 fit into Bittensor? Bittensor’s official subnet page says Subnet 102 implements a distributed expert architecture on the network. (connito.ai) The page says miners run expert subnetworks independently, while validators coordinate expert updates and assign rewards based on performance metrics. Connito’s GitHub repository uses similar language. Its README says the subnet is built for collaborative decentralized training of large language models with a Mixture-of-Experts architecture, and says the model is split into expert groups distributed across many independent miners. (connito.ai) ### Why does Mixture-of-Experts keep coming up? Connito’s architecture page says MoE activates only a subset of model parameters for a given input, which it says allows model capacity to scale without compute rising at the same rate. (bittensor.ai) The company also says individual domain experts can be updated without retraining the whole system. The Bittensor subnet page frames the same point in operational terms. It says distributing model training across independent expert groups lowers the compute burden on each node and lets organizations contribute heterogeneous hardware. (github.com) ### What does “training-as-a-service” appear to mean here? Connito’s public materials describe domain-specific training for customer use cases rather than a general chatbot launch. Its explainer pages cite legal, medical and technical experts as examples of specialized model components, and say organizations can maintain control over outputs while training domain experts across distributed teams. (connito.ai) The company’s whitepaper says Connito combines sparse MoE training with what it calls a “Proof-of-Loss” incentive mechanism, in which workers submit updated experts and validators test those updates on held-out data before useful changes are integrated into the global model. (bittensor.ai) ### What should people watch next? May 26, 2026, is the next concrete date attached to the project in the reviewed material. That is the date around which the May 22 thread said a dashboard and early beta access would appear, while Connito’s website and documentation continue to describe the underlying subnet, miner and validator system. (bittensor.ai) (connito.ai 1) (connito.ai 2)