Bittensor Drama! TAO down 15%! | E2274

Key Concepts

• BitTensor (TAO): A decentralized network for machine learning models where participants (miners) compete to provide services, incentivized by the TAO token.
• Subnets: Specialized, modular networks within BitTensor (128 total) dedicated to specific tasks like video processing, coding, or AI training.
• Rug Pull: A situation where a project founder abandons a project and potentially misappropriates funds or tokens, causing a sharp decline in value.
• Permissionless Participation: The ability for anyone globally to contribute compute or AI models to the network without needing centralized approval.
• LLM Council: A framework (inspired by Andrej Karpathy) where multiple AI personas debate a problem, peer-review each other anonymously, and a "chairman" model synthesizes the final decision.
• SML (Small Model Languages): Efficient, specialized AI models that can run locally on hardware, often outperforming massive general-purpose models for specific tasks.

---

1. The BitTensor "Rug Pull" Incident

The show addressed a recent controversy involving Covenant AI, which operated three subnets (3, 39, and 81) focused on decentralized AI training.

• The Allegations: Sam Dar, the lead of Covenant AI, was accused of selling off his TAO tokens and abandoning the subnets. Covenant AI claimed that BitTensor co-founder Jacob Steves (known as "const") obstructed their operations, leading to the defection.
• The Impact: TAO’s price dropped from ~$335 to ~$271.
• Expert Perspective: Guest Gareth Howles (Video Subnet 85) noted that while Covenant AI was a "marquee" project, the BitTensor ecosystem is resilient. The consensus is that this is a growing pain of a decentralized system, highlighting the need for better subnet governance and staking requirements to prevent future "rug pulls."

2. Video Subnet 85: Real-World Application

Gareth Howles discussed Subnet 85, which focuses on video optimization.

• Functionality: The subnet uses AI agents to perform video upscaling (e.g., VHS to 4K), colorization, frame rate adjustment, and metadata tagging.
• Market Opportunity: With 85% of internet traffic being video, there is massive demand from archives (BBC, Getty Images) and consumers to modernize legacy content.
• Economic Model: The subnet leverages a "winner-takes-most" model where miners compete to provide the best model outputs. This creates a global, permissionless labor market where developers from regions like Vietnam can contribute high-level ML work without traditional corporate barriers.

3. Framework: The "Council of Advisers"

Guest Ola Layman demonstrated a tool built for Claude that automates complex decision-making.

• Methodology:
  1. Persona Generation: The system creates distinct personas (e.g., Contrarian, Expansionist, First-Principles Thinker, Executor).
  2. Parallel Processing: Each persona generates an answer to a prompt.
  3. Anonymous Peer Review: Personas review each other’s work without knowing who wrote which response, eliminating bias.
  4. Chairman Synthesis: A final model reviews the debate and provides a consolidated recommendation.
• Application: This is highly effective for startup founders (e.g., determining equity splits for a new hire) as it saves "biological compute" and provides a balanced, multi-perspective analysis.

4. Key Arguments & Perspectives

• Capitalism vs. Regulation: Jason Calacanis argues that decentralized networks like BitTensor represent "capitalism at its finest," allowing global competition to drive down costs and improve service quality, bypassing traditional corporate and geographic constraints.
• Investment Strategy: Calacanis views buying TAO as an "ETF of startups," providing exposure to the entire BitTensor ecosystem rather than betting on individual subnets.
• Security Concerns: Ola Layman described the current state of AI security as a "holding pattern," urging founders to take basic security measures (like managing API permissions) seriously, even if the long-term risks of advanced AI remain unpredictable.

5. Notable Quotes

• Jason Calacanis: "My job is to learn by betting and to have conversations... the way I choose to place bets is not by writing code, not by being a subnet participant... it is studying, placing bets, losing bets, learning, placing more bets."
• Ola Layman: "Most people think AI is ChatGPT... [but] if you have to think and empathize with a different opinion every day, it takes so much of your own biological compute."

6. Synthesis/Conclusion

The episode highlights a pivotal moment for decentralized AI. While the Covenant AI incident exposed vulnerabilities in subnet governance, the underlying thesis—that permissionless, global competition can produce superior AI services—remains strong. The shift toward "SMLs" and frameworks like the "Council of Advisers" suggests that the next phase of AI development will be defined by specialized, efficient, and highly collaborative tools that empower individual founders and developers to operate at a scale previously reserved for large tech corporations.