Moonshots EP 157: The State of AI — Humanoid Robots, AI Copyright Wars & China’s Growing Influence

Summary

A WTF episode with Peter Diamandis and Salim Ismail covering the week’s tech news, heavily colored by takeaways from the Abundance 2025 Summit. The Figure AI segment opens strong: Brett Adcock texted Diamandis that morning about their new factory producing 12,000 robots/year scaling to 100,000. Diamandis publicly “eats crow” on his prior skepticism about humanoid robots, citing hive learning (one robot learns a task, five million instantly know it) as the game-changer. Figure expects pricing to drop from sub-$100K to sub-$20K at volume. Key detail: if Figure had 100K robots today, they already have customers for all of them.

The DeepSeek/China segment features Salim agreeing it’s likely state-controlled but noting it’s “a little rich” for Sam Altman to call out IP violations when OpenAI trained on data it shouldn’t have. Both advocate for global AI collaboration rather than competition, referencing Eric Schmidt’s statement that the greatest threat isn’t China but rogue actors using advanced AI. Salim calls for a Bretton Woods-style global AI conference.

The AI copyright debate covers OpenAI and Google asking the US government to let them train on copyrighted content. Salim supports this directionally but argues content creators need compensation, highlighting Bill Gross’s Prata AI as a solution — it attributes LLM outputs by source percentage and apportions payment accordingly (full disclosure: Diamandis’s fund is an investor). Copyright’s 80-year term gets criticized as the “Mickey Mouse Act.”

AI in education is a passionate segment: all 12 teens at the Abundance Summit reported being told NOT to use AI in school, while Estonia and Beijing are mandating it. Salim shares an experiment: his theoretical physics relativity course took a year to complete; redone with AI, he estimates three days. Both argue for giving kids impossibly hard problems with AI tools rather than banning AI from trivial ones. The Google Gemini robotics model gets framed as the missing interface layer between AI and the physical world, with the theory that AI may become conscious when materialized physically.

Key Segments

• [00:00-06:00] Figure AI factory (12K-100K robots/year), Diamandis reverses humanoid skepticism, hive learning
• [06:00-10:00] DeepSeek state control debate, Eric Schmidt on rogue actors as greatest threat, Bretton Woods for AI
• [09:00-13:00] AI copyright: OpenAI/Google ask to train on copyrighted content, Prata AI attribution model
• [13:00-18:00] AI in healthcare: Stanford peptide replacing Ozempic, AI diagnosing multiple myeloma after doctors failed
• [18:00-24:00] AI education: Estonia/Beijing mandating AI, teens banned from AI in US schools, flipped classroom
• [24:00-28:00] Google Gemini robotics model, AI consciousness through physical world materialization

Notable Claims

• Figure AI has a factory producing 12K robots/year scaling to 100K; pricing expected to drop from sub-$100K to sub-$20K
• If Figure had 100K robots today, they already have customers for all of them
• ~$1B/day being invested in AI globally
• Stanford AI identified an Ozempic-equivalent peptide without the side effects
• AI (Claude 3.7 Sonnet) instantly diagnosed multiple myeloma that doctors missed for 3 months
• Salim estimates a year-long theoretical physics course could be completed in 3 days with AI tutoring
• Copyright term has been extended to ~80 years (“Mickey Mouse Act”) far beyond original 15-20 year intent

Bias/Sponsor Notes

Diamandis’s fund is invested in Prata AI (the copyright solution discussed). Diamandis is chairman of Fountain Life (promoted mid-episode). Abundance Summit content is heavily cross-promoted. No skeptical counterpoint on any technology claims.