September was defined by subnets. The Opentensor team spent the entire month preparing for the October 2 launch and we are very proud to have executed a smooth one. Subnets pave the way for a significantly enhanced AI developer experience by offering access to valuable resources such as storage and compute. We are optimizing for real world value and over the coming months, we will be spending the majority of our time helping developers build subnets that others can use. If we are successful, early developers will be rewarded, their users will have a cheaper and faster product, and all holders of TAO will have access to an increasingly useful system, one that has the potential to become the infrastructure for all AI development.
We are currently making decisions for the next ten years, constantly asking ourselves “what can we do today to create a market that incentivizes the production of products that have enduring value?” We are building out robust metrics to track real subnet usage, and will be working very closely with subnet owners to maximize the odds of their, and thus the network’s, success.
Metrics are hyper-specific to each subnet. For example, on a compute subnet, we will need to constantly monitor the cost of compute. While on a text-based machine intelligence subnet, metrics will be slightly more complex, with both speed and quality of response being critical. Subnets will master only what they measure — thus, we encourage subnet owners to track a few important metrics as opposed to tens of irrelevant ones. Focus on signal, not noise. By doing so, subnets will be able to craft a more precise validation mechanism which will result in higher quality miner outputs, which will result in our ultimate goal: end-user satisfaction.
For newer readers, you may be asking: what are subnets? The below is pulled from this X thread and is helpful for a high-level understanding. For those wishing to dive deeper, we highly recommend the Bittensor Paradigm.
The unveiling of user-created subnets represents a historic moment for both Bittensor and the future of decentralized intelligence. The Bittensor language is being opened to the world, allowing any individual to create markets for the commodities needed to create intelligence.
A decentralized commodity market is powerful for the same reason Google, OpenAI or any tech giant is powerful. Intelligence is the product of several digital commodities (compute, data, etc.) and until now, these commodities have only been available under siloed corporate roofs.
The advent of the user-created subnet will unleash a cambrian explosion of digital commodity markets, all under one unified token system. Now, any developer will have access to the same resources as those historically privileged few who operate within the closed doors of Big Tech.
The only difference is: on Bittensor, these resources will be more powerful, as our network leverages compute and expertise from every corner of Earth. By directing the power of digital markets towards society’s most important digital commodity — AI — we will build the most powerful intelligence network in the world.
And more importantly, by enabling global participation and ownership, Bittensor will ensure that AI remains in the hands of the people. We envision a future where we are all equal and all possess control over the most powerful technology humanity has ever created.
There has been significant organic interest in subnets, with three already registered this week:
We are currently working with several other developer teams who are progressing with subnet ideas of their own. Over the coming weeks, we expect additional subnets to be registered, with validators (and thus miners) beginning to move over to the ideas they believe in most. It is the first inning of the subnet wars.
The team had a bit of travel this month as well, with Ala and James traveling to Singapore for Token2049 and Jake to New York for Mainnet, where he spoke alongside Akash and Near. Both conferences were highly productive, filled with meetings with many key stakeholders. The consensus from both conferences: Bittensor is on the bleeding edge of both blockchain and AI and has an incredible opportunity to revolutionize both. We don’t intend on wasting the opportunity to do so.
Moving forward, we plan to spend less resources traveling to crypto conferences (likely limiting to just 1–2 a year) and will be having a bigger presence at events where traditional AI/ML developers are present. Ultimately, our network is only as valuable as the outputs generated, and the best developers tend to produce the best outputs. The reason to develop on Bittensor is clear: we have created a market where you, the individual, are rewarded directly for the quality of your contributions. The incentive is strong and we expect competition to increase as more of the world begins to understand the promise and potential of Bittensor.
On the theme of developers, we have added two technical writers to our team this month: Raj and Dan. Both have experience in AI and blockchain, with Raj previously writing for Cerebras and Aptos and Dan producing content for Axelar, Alchemy and a stealth AI startup. We have added both writers to the team as we want to significantly enhance the developer experience. Step one in doing so is improving our existing documentation. People tend to do more of what’s convenient and friction-free and developers are no different. Understanding this, we intend to drastically improve the Bittensor developer experience, making it as simple as possible. Admittedly, this has not been a historical point of emphasis for us, but we recognize the need to change. This will not happen overnight, but we are not making decisions for overnight — we are making decisions for the long term. As always, we welcome any and all feedback from developers, no matter how harsh.
We are incredibly excited about how far we have come but more importantly, where we are headed. We guarantee you there is more innovation ahead of us than behind us, and we thank you as always for your support — we have the best community not just in frontier technology, but in the world.