Novelty Search August 7, 2025

This episode reveals how two pioneering Bittensor subnets are tackling crypto's incentive problems and AI's scaling limitations, offering a blueprint for sustainable, decentralized innovation.

Introduction: The Architects of Bittensor's Core

Host Jake from OpenTensor introduces the episode's key speakers: Cameron (alias "Vune"), Joseph (JJ), and Abe of Leighton Holdings. Jake provides context on their deep involvement, highlighting Cameron's journey from a pivotal intern to a core maintainer of the Bittensor ecosystem. He emphasizes that Leighton Holdings stepped in to fund the critical, and often thankless, work of maintaining Bittensor's core infrastructure, including the SDK and API, after the OpenTensor Foundation began decentralizing its operations.

• Speaker Context: Jake frames the speakers not just as developers but as long-term, altruistic contributors who have been essential to Bittensor's stability and growth.
• Key Quote (Jake): "He has fixed the chain and saved all of us from being destitute a number of times... Cameron, just about runs the Nucleus team at OpenTensor, even though he's not at OpenTensor."

Leighton Holdings: Powering Bittensor's Open-Source Engine

Joseph (JJ) outlines Leighton Holdings' mission, stating it is synonymous with the Bittensor mission. Operating for about nine months, the small, mission-driven team focuses on making Bittensor accessible, stable, and understandable for developers and users.

• Core Contributions:
  • Tooling & Infrastructure: Maintenance of the Bittensor SDK, CLI, API server, wallet, and other core open-source tools.
  • Analytics & Education: Development of Tao.app, an explorer and analytics platform, and Learn Bittensor, an educational resource to demystify the protocol.
  • Open-Source Impact: In the last six months, Leighton has produced over 100 releases, nearly 4,000 code commits, and almost 1,000 pull requests, driven by a core team of just four engineers.
• Strategic Insight: Leighton's work demonstrates the vital role of dedicated infrastructure teams in a decentralized ecosystem. For investors, the stability and quality of this core tooling are foundational to the network's long-term value and ability to attract developers.

Taoash (Subnet 14): A Decentralized Answer to Bitcoin Mining Centralization

Abe introduces Taoash, a Bittensor subnet designed to operate as an openly owned and accessible Bitcoin mining pool. The project's primary goal is to combat the centralization in Bitcoin mining, where just three pools control over 56% of the network's hashrate—the total computational power dedicated to mining.

• How It Works:
  1. Miners direct their hashrate to the Taoash pool.
  2. The pool contributes this hashrate to the Bitcoin network and earns Bitcoin rewards.
  3. Bitcoin is passed back to the miners, while Taoash incentivizes them with its native token, TAO (referred to as "alpha" in the discussion).
  4. A small pool fee is collected and distributed to stakeholders in the subnet.

The Original Sin: Learning from Taoash's Initial Design Flaws

Jake prompts a discussion on the "drama" surrounding Taoash's initial launch. Abe provides a transparent breakdown of the original design and its critical failures, offering a case study in decentralized economic design.

• The Flawed Premise: The original model attempted to subsidize 100% of the Bitcoin hashrate's value with TAO emissions. Validators received the mined Bitcoin and were trusted to use it to buy back TAO, driving value back to the subnet.
• Key Problems Identified:
  • Scalability Failure: The subnet's maximum supportable hashrate was directly tied to the TAO price. Any price dip immediately reduced the hashrate it could incentivize, creating a risk of a "downward spiral."
  • Misaligned Incentives: Validators had no strict obligation to return the mined Bitcoin. Many simply kept it to fund their own operations, breaking the economic loop. Abe candidly calls it a "horrible design looking back."
  • Inefficient Buybacks: The buyback mechanism was vulnerable to MEV (Maximal Extractable Value), where traders could front-run the purchases, diminishing their impact on returning value to token holders.
  • Currency Risk: The model was highly sensitive to the TAO-to-Bitcoin price ratio, creating instability.

The Taoash Redesign: A Sustainable Path Forward

The team redesigned Taoash to function more like a traditional, competitive mining pool, using TAO as a targeted subsidy rather than a full replacement for Bitcoin rewards.

• The New Model:
  • Miners receive their mined Bitcoin directly, minus a low 2% pool fee.
  • TAO emissions now act as a subsidy, effectively lowering the miners' net pool fee to a highly competitive rate (targeting 1.5-1.75%).
  • This decouples the subnet's viability from the daily TAO price, allowing it to scale its supported hashrate linearly with the TAO price, up to a theoretical maximum of 7% of the entire Bitcoin network.
• Actionable Insight: This redesign is a crucial lesson for Crypto AI investors. Subnets with flawed economic models that create negative feedback loops are unsustainable. The shift to a subsidy model makes Taoash more resilient and competitive, a key factor to consider when evaluating its long-term potential.

The Future of Taoash: A Trustless Hashrate Marketplace

Abe outlines the roadmap, which focuses on further decentralization and creating new value streams from the aggregated hashrate.

• EVM-Based Hashrate Marketplace: The flagship future project is a trustless marketplace built on an EVM (Ethereum Virtual Machine) compatible chain. This will allow the subnet to sell its aggregated hashrate to the highest bidder, creating a premium revenue stream beyond standard mining. This is modeled after successful centralized services like NiceHash but aims to be fully decentralized and permissionless.
• Decentralized Custody & Infrastructure: Future plans include using smart contracts to custody mined Bitcoin and having validators run pool infrastructure, progressively removing all single points of failure and trust from the system.

Subnet 5: A New Frontier in Hierarchical AI Training

Roberto (Rob) from Manifold announces a new collaboration with Leighton Holdings on Subnet 5, a project named Hone. Its mission is to pioneer a new path to AGI by focusing on hierarchical learning to solve the notoriously difficult ARC-AGI-2 benchmark, where current large models stagnate at around 5% accuracy.

• The Vision: The project moves away from the "scale is all you need" philosophy. Instead, it leverages architectures like JEPA (Joint-Embedding Predictive Architecture), which aim to build efficient world models that learn and reason in levels, much like the human brain. This approach promises greater sample efficiency and the ability to solve novel problems with minimal examples.
• How It Works:
  1. The subnet will focus on training relatively small but powerful hierarchical models (e.g., 1-10 billion parameters).
  2. Miners will contribute by producing and sharing gradients in a distributed training process.
  3. The goal is not to create one massive model but to create an "AI model factory" that can rapidly iterate and improve to achieve a high score on the ARC-AGI-2 benchmark.

A New Paradigm: Benchmark-Driven, Capital-Efficient AI

Jake raises a key question: if the models are small, why use a distributed network? Rob explains that the strategy is not about aggregating massive compute but about leveraging Bittensor's incentive mechanism to find the most cost-efficient path to solving the benchmark.

• Strategic Advantage: The ARC-AGI-2 competition has a limited compute budget, making brute-force scaling impossible. Subnet 5 aims to use Bittensor's economic incentives to optimize training costs, turning the network into a tool for rapid, capital-efficient R&D.
• Actionable Insight for Researchers: This project signals a new application for decentralized AI networks: targeted, benchmark-driven research. Instead of competing with hyperscalers on model size, researchers can use Bittensor to innovate on architectural efficiency and solve specific, high-value problems.
• Timeline: Rob commits to launching a testnet and initial version of the "AI factory" within two weeks, inviting community participation.

Conclusion: From Raw Incentives to Strategic Innovation

This episode highlights Bittensor's evolution into a sophisticated platform for both crypto-economic design and cutting-edge AI research. The Taoash redesign and the launch of Hone show a move toward sustainable, benchmark-driven subnets. Investors and researchers should monitor these models as blueprints for future value creation on the network.