Distributed training has never been easy. Building a distributed system that makes economic sense on Bittensor is harder still. Today, we’re introducing Connito: a network for collaboratively building composable specialized AI. Signal shows us that just by training a particular selection of experts as a partial model, we can improve the performance of the whole system. Learn about our architecture and results: connito.ai/blog/distributing…

Frontier-scale training has belonged to labs with giant GPU clusters. @ConnitoAI is trying to break that model. At Exploit, @isabella618033 shows how we can turn distributed training model into a business model that we can sell to scale: luma.com/exploitsummit26

We're getting ready to launch our training-as-a-service (TaaS) product! And we want to build it around what you want. This initiative would act as our market research, done in the open, tell us which expert you wish existed, and the Connito subnet may train it for real. Every proposal, every 👍 and every result shapes the platform our customers will use to train their own experts. connito.ai/

Yes, forgetting happens when you teach a model new things. But forgetting dose not just happen even through out the model. Research says that it happens particularly in the router. To become an insider for how LLM works: connito.ai/blog/why-does-fin…

Finetuning Instead of Expensive Frontier Models Finetuning gives you opportunity to build smaller and more efficient models that can fit into your at home GPU. Buttt at the cost possibly forgetting old knowledge. Check out this explanatory post of how do models learn new things! connito.ai/blog/why-does-fin…

Building Modular Intelligence Would you prefer hosting AI model on your own infrastructure or use direct API access? McKinsey's report Open source technology in the age of AI indicate that most prefer the first. How is it the case? And how can Connito make it easier for custom users? Read more: connito.ai/blog/one-giant-mo…

Building Modular Intelligence Ever wondered how can model inference be cheap for providers, but expensive when you try to run it locally? The answer could be architectural due to model parts utilization. Check out what Connito offers to make this effort smooth for you. Read more: connito.ai/blog/one-giant-mo…

--- Building Modular Intelligence Series --- What if the future of AI isn’t one model that knows everything? For many tasks, we don’t need every capability a giant model carries. We need the right capability, at the right level of performance, without paying for everything else. In Distilling Step-by-Step, a 770M-parameter T5 model outperformed few-shot 540B-parameter PaLM on a defined task. More recently, Inkling-Small also beat the much larger Inkling on several reasoning and agentic coding benchmarks. Check out our latest post explores the case for model specialization 🧩 connito.ai/blog/one-giant-mo…

Excited the see the @bittensor community at the @ExploitSummit Sept 28-29th, 2026 Montreal, Canada

Over the past several days, we've been working around the clock to understand the security incident that impacted the ORO subnet owner wallet on July 13th. First, the most important thing: the subnet is fully operational. The breach was contained to only the SN15 owner wallet, which had 147K Alpha Tokens transferred to the attacker. The team has worked closely with the @opentensor, @CrucibleLabs, @ConnitoAI, Crypto Exchanges, Law Enforcement, and members of the community to identify the root cause and have filed reports with the necessary authorities in an attempt to recover some of the funds that were stolen. We believe that the source of the breach was a team member’s computer that had been compromised and exfiltrated the owner wallet. The team is still gathering information to determine exactly how this happened. We do have evidence pointing to a single attacker behind both our and Connito's incidents. There is no evidence that the Crucible Wallet is the source of this attack. When we have more information, we will share the full post-mortem with the community to ensure this doesn't happen to anyone else in Bittensor. When we bought the subnet slot, we'd decided both the validator and the owner wallet would sit on hardware wallets. In a coordination lapse, the owner key was set up as a software wallet instead. That was inexcusable, and it was our mistake. We are truly sorry for the impact this has had on our community and our supporters. This incident reinforced an important lesson for the team: building great products also requires continuously evolving our operational security. As part of that commitment, we've already implemented or are implementing: • A coldkey swap of SN15 ownership. • Multi-signature hardware protection for the subnet owner keys. • A Conviction lock on owner emissions. • Additional internal operational security controls and reviews. Our priority now is rebuilding the community's trust and continuing to ship improvements to the subnet. For those who have stuck by our side, we thank you for your continued support. The best is yet to come. Yours, ORO

Recovered my personal account, feel free to follow

Hello all, Apologies for the delay as I've just woken up to this. We are still here and not going anywhere. We’ve identified that ~2500 TAO was transferred out of the subnet owner account, and my personal twitter account, @isabella618033, has been deleted. We are still investigating what happened and working to understand the full scope of the issue. As mentioned we are still here and will continue building while we work through this. Stay tuned for updates as we figure this out. Thanks for your support

AI is moving from one giant model to a library of reusable skills. You can now upgrade what an AI is good at without rebuilding the whole thing. Coding, math, safety and any customer's task. Here's why that changes everything. connito.ai/blog/modular-mode…

Inside our decentralized training breakthrough: using 5× less memory than the full model while training 4× faster than ESFT. connito.ai/blog/subnet-statu…

The future of AI is when every household computer can help train decentralized intelligence. Open, distributed, and powered by everyone.