This episode reveals how Bittensor's top builders are pivoting from early-stage incentive models to sophisticated, sustainable subnet economies, tackling everything from Bitcoin mining decentralization to the next frontier of AGI research.
Introduction to the Team and Latent Holdings
The episode brings together key figures who have been instrumental in Bittensor's development. Jake, the host, introduces Cameron (aka Vune), a core developer who has maintained critical infrastructure like the bittensor API, and Abe, another long-term contributor. Both are now central to Latent Holdings, a team that took on the responsibility of maintaining core Bittensor tooling after the OpenTensor Foundation began decentralizing its operations.
Joseph, representing Latent Holdings, outlines the team's mission: to make Bittensor more accessible, stable, and understandable. Their work is foundational to the ecosystem's health.
- Core Contributions: Latent Holdings maintains the Bittensor SDK, CLI, developer documentation, and other open-source tools. They are the largest open-source contributor to Bittensor outside the foundation.
- Key Projects: The team is behind community-facing projects like Tao App (an analytics explorer) and Learn Bittensor (an educational resource).
- Impressive Velocity: In just six months, the small team has produced over 100 releases, nearly 4,000 code commits, and changed over 173,000 lines of code, demonstrating immense dedication and efficiency.
TaoHash (Subnet 14): Decentralizing Bitcoin Mining
Abe introduces TaoHash (Subnet 14), a project designed to challenge the centralization in Bitcoin mining. He highlights a critical vulnerability in the Bitcoin network: just three mining pools control a majority (56.8%) of the network's hash rate. TaoHash aims to solve this by creating an openly owned and accessible Bitcoin mining pool.
- Hash Rate: This is the total combined computational power being used to mine and process transactions on a proof-of-work blockchain like Bitcoin. Higher hash rate means greater network security.
- How it Works: Miners direct their hash rate to the TaoHash pool. The pool contributes this power to the Bitcoin network, earns Bitcoin rewards, and passes them back to the miners. The subnet's native token, Alpha, is used to provide an additional incentive layer.
- Value Proposition: TaoHash offers miners one of the lowest pool fees on the market (around 2%), lightning-fast payouts, and full transparency with open-source code.
Abe states the core problem simply: "56.8% of the Bitcoin network hash rate is controlled by just three pools. This is a big problem and it's not good for the Bitcoin network."
The Evolution of TaoHash: Learning from Past Mistakes
The conversation turns to the "drama" surrounding TaoHash's initial design. Abe provides a candid post-mortem on the original model and the critical lessons learned, offering a valuable case study in subnet economics for investors and researchers.
The old design had several fundamental flaws:
- Scalability Issues: The subnet's ability to incentivize hash rate was directly capped by the price of its Alpha token. A drop in the token price would immediately reduce the hash rate the subnet could support, creating a risk of a "downward spiral."
- Misaligned Validator Incentives: The original model trusted validators to sell the mined Bitcoin and use the proceeds to buy back Alpha, driving value back to the subnet. Instead, many validators simply kept the Bitcoin, breaking the economic loop.
- Inefficient Buybacks: The buyback mechanism was susceptible to MEV (Maximal Extractable Value) and failed to distribute value evenly to all token holders.
- Currency Risk: The subnet was exposed to volatility in the TAO-to-Bitcoin conversion rate.
The Redesigned TaoHash: A More Sustainable Model
The team redesigned TaoHash to function more like a traditional mining pool, but with a crypto-economic twist. This new model is more robust and sustainable.
- Direct Bitcoin Payouts: Instead of trying to capture 100% of the mining rewards, TaoHash now passes the majority of the mined Bitcoin directly back to the miners, minus a small pool fee.
- Alpha as a Subsidy: The Alpha token now acts as a subsidy, effectively lowering the net pool fee for miners to ~1.5-1.75%, making it highly competitive. This removes the direct dependency on the Alpha price for incentivizing hash rate.
- Strategic Implication: This pivot demonstrates a maturing understanding of subnet economics. Investors should favor models that provide direct, tangible value (like BTC payouts) and use native tokens for subsidies or governance, rather than as the sole, volatile reward mechanism.
Future Roadmap: A Trustless Hash Rate Marketplace
TaoHash's ambition extends beyond being just a mining pool. The team is building an EVM-based, decentralized hash rate marketplace, inspired by services like NiceHash.
- Hash Rate Marketplace: This will allow the subnet's hash rate to be sold trustlessly to the highest bidder. This financializes hash rate, turning it into a tradable commodity.
- Benefits: This creates a premium on the hash rate, providing faster payouts for miners and higher returns for Alpha token holders who govern the market. It also removes the team as a trusted intermediary.
- Decentralized Infrastructure: The long-term vision includes decentralizing the pool infrastructure itself, likely having validators host endpoints to create a fault-tolerant, distributed system.
Subnet 5 (Hone): A New Frontier in Hierarchical AI
The discussion shifts to a new collaboration between Latent Holdings and Manifold on Subnet 5, codenamed "Hone." Rob from Manifold passionately outlines an ambitious mission: to achieve a breakthrough in Artificial General Intelligence (AGI) by focusing on novel architectures rather than just scaling up existing models.
- The Goal: The project's first concrete objective is to solve the ARC-AGI-2 benchmark, a notoriously difficult test for current AI models, where even state-of-the-art systems like GPT-4 stagnate at around 5% accuracy.
- The Approach: The subnet will focus on training hierarchical models, inspired by concepts like Yann LeCun's JEPA (Joint-Embedding Predictive Architecture). These models learn abstract representations of the world, enabling them to reason and plan in a single forward pass, unlike the thousands of passes required by current auto-regressive models.
- Key Insight: This approach promises extreme sample efficiency. A recent paper showed a small, 27-million-parameter hierarchical model achieved 6% on ARC-AGI-2 with only 1,000 training examples—outperforming models thousands of times larger.
Rob frames the mission with urgency and clarity: "We're going to be contributing to open source AI by solving the ARC-AGI-2 benchmark, which no one has solved yet."
Mechanics and Strategic Implications of Subnet 5
The subnet will function as a distributed training network, but with a unique focus.
- Smaller, Efficient Models: Unlike other training subnets aiming for massive parameter counts, Subnet 5 will train smaller models (starting in the 50-100 million parameter range). This makes participation accessible to researchers with consumer-grade hardware (e.g., a single 4090).
- Rapid, Cost-Effective Iteration: The combination of small models, high sample efficiency, and a clear benchmark allows for rapid, low-cost innovation. The goal is to find a solution that is not only intelligent but also computationally cheap, a requirement of the ARC-AGI-2 prize.
- Actionable Insight for Researchers: This project signals a shift away from the "scale is all you need" paradigm. Researchers should watch Subnet 5's progress closely, as a breakthrough would validate the potential of hierarchical architectures and open new avenues for AGI research beyond brute-force scaling.
- Timeline: Rob commits to a public timeline of "two weeks" for launching the initial testnet, signaling a fast-moving and highly focused effort.
Conclusion
This episode highlights the maturation of the Bittensor ecosystem, showcasing a move towards sustainable tokenomics with TaoHash and a pivot to cutting-edge, targeted research with Subnet 5. For investors and researchers, the key takeaway is to prioritize subnets with robust economic designs and novel technical approaches that promise step-function progress over incremental gains.
