← All episodes

Bittensor’s (alleged) $10M rug pull (feat. Mark Jeffrey) | E2275

| 4 products mentioned
Watch on YouTube decentralized ai bittensor subnets crypto incentives deepfake detection distributed computing blockchain governance early-stage investing

This episode explores Bittensor, a decentralized AI network attempting to democratize frontier model training, but is dominated by discussion of a major ecosystem failure: subnet operator Sam Dair's alleged $10M rugpull of his Templar project. Mark Jeffrey, an early Bitcoin adopter and crypto veteran, breaks down why the incident happened despite Bittensor's elegant incentive-alignment architecture, and what structural fixes (like token lockups and ownership mechanisms) could prevent future betrayals—making this essential viewing for anyone considering participation in early decentralized AI infrastructure.

Key takeaways
  • Bittensor's incentive-alignment engine works "spectacularly well," but collapsed under sudden success: when subnet operators accumulate tokens and liquidity pools exist, they can execute a rugpull by dumping tokens to Uniswap and converting to fiat, as allegedly happened with Templar's $37,000 TAO (~$10M).
  • Token lockups and conviction-based ownership are the structural fix: proposals now require subnet operators to lock their tokens for extended periods, with ownership going to whoever locks the most tokens longest—forcing founders to prove commitment or lose control.
  • Decentralized AI (like Bittensor) offers retail investors access to early-stage AI companies they'd normally be blocked from, bypassing accredited-investor gatekeeping that keeps 90% of upside with the wealthy.
  • IOTA's "train at home" model fractures expensive GPU training into 20-minute MacBook-sized compute chunks: this makes frontier model training economically feasible by aggregating idle consumer compute and interruptible cloud instances at 10 cents on the dollar.
  • BitMind uses competitive red-team/blue-team mining to detect deepfakes and AI-generated content at scale, with miners competing to build the best detection models while adversaries try to fool them—creating a continuously hardening system that updates faster than centralized approaches.
  • Authenticity and meta-commentary (not spectacle) can drive engagement: Justin Bieber's minimalist Coachella set—performing over his own YouTube videos—resonated because it told a coherent story about his career, whereas pure showmanship without narrative depth ages poorly.
  • Proper exit negotiation (not rugpulls) means a founder keeps some tokens, hands off the subnet key to the foundation, and lets the community continue, aligning incentives and building long-term credibility in the ecosystem.

Recommendations (2)

Zebra G750 uses

"This is my um Zebra G750. But as fast as we're talking, I'm not going to get it all down right. I might write down two or three words."

This Week in Startups · ▶ 5:44

BitTensor
BitTensor uses

"When I encountered Bittensor and started digging around the subnets, having been a participant in many early ecosystems, what I saw in Bittensor was quite a lot of early signal."

Mark Jeffrey · ▶ 3:08

Mentioned (2)

Uniswap
Uniswap "Along came Uniswap and they said why don't we do something like a Coinbase or a Kraken or a Binan..." ▶ 4:45
Cursor
Cursor "Sentry works alongside coding agents like Cursor, passing along an idea for a fix that can be app..." ▶ 9:57