Did OpenAI Steal a Millennium Proof From a Private AI Chat? - Uneasy Money

Summary

The episode covers the Liquid Network sidechain hack, where a consensus bug allowed the minting of unbacked LBTC and the theft of ~4,000 BTC, followed by a partial return and extortion attempt. It then discusses two new frontier AI models, Astra and Fable 5.1, and the controversy over OpenAI allegedly training on private math proofs. The conversation explores AI privacy, open-source models, identity, and the incentive problems driving the race to superintelligence.

• Liquid Network suffered a consensus bug allowing unbacked LBTC minting and BTC drain.
• Hackers returned 85% of funds but are extorting Blockstream for the rest.
• Discussion of white-hat claims and bad incentives in crypto hacking.
• Two new AI models, Astra and Fable 5.1, were released.
• OpenAI faced controversy over allegedly using private AI chats to solve a math proof.
• Privacy concerns rise as user data may end up in training sets despite toggles.
• Open-source models and private inference are presented as alternatives.
• Debate on AI safety, alignment, and incentive-driven races among frontier labs.