The State of AI: Humanoid Robots, AI Copyright Wars & China’s Growing Influence w/ Salim Ismail #157
Peter Diamandis and Salim Ismail run their weekly news roundup plus highlights from the 2025 Abundance360 Summit. They cover Figure AI's humanoid robot factory, the Sam Altman/DeepSeek China fight, the OpenAI/Google push to train on copyrighted content (and Bill Gross's Prorata AI revenue-sharing fix), AI accelerating medical diagnosis and longevity, Estonia and Beijing mandating AI in schools while Western classrooms ban it, Ilya Sutskever's Safe Superintelligence, and Max Hodak's biohybrid brain-computer interface at Science Corp. They close on geopolitical risk, the dystopia-before-abundance framing borrowed from Mo Gawdat, and recaps of Palmer Luckey and the AI-investing panel from the summit.
Figure AI's new factory can build 12,000 robots/year, scaling to 100,000/year; Diamandis says he was wrong to be skeptical of humanoid robots given the pace of iteration.
Sam Altman vs. DeepSeek and ChinaGeopolitics▶ 6:28
Altman calls DeepSeek state-controlled and wants Chinese models banned; Ismail agrees it's likely true but calls it hypocritical given how OpenAI itself sourced training data, and argues the real threat is rogue actors, not China.
AI copyright fight and the Prorata fixEconomy▶ 9:26
OpenAI and Google are asking the US government to let them train on copyrighted content; Diamandis argues copyright law has been twisted to serve corporations, while Ismail points to Bill Gross's Prorata AI as a model that apportions LLM revenue back to the original content sources.
AI-accelerated diagnosis and longevityLongevity▶ 13:41
Stanford researchers used AI to find a peptide that mimics Ozempic's weight-loss effect without the side effects; Diamandis recounts AI catching a multiple myeloma diagnosis that doctors missed for three months, and says doctors misdiagnose roughly 30% of the time.
AI mandated in schools: Estonia and BeijingOther▶ 18:23
Estonia is rolling out ChatGPT for all secondary schools and Beijing is mandating AI courses in primary and secondary schools, while most Western classrooms Ismail surveyed still ban AI use outright.
Google Gemini Robotics bridges AI and the physical worldRobotics▶ 25:12
Google's Gemini Robotics model lets one AI software layer control many different physical robot bodies; both hosts frame physical embodiment as the 'Holy Grail' and note theories that consciousness emerges from interacting with the physical world.
Ilya Sutskever's Safe Superintelligence and defining ASIAI▶ 27:46
Sutskever's startup is raising at a $30B valuation around the term 'superintelligence'; Ismail pushes back that neither AGI nor superintelligence has an agreed definition, while Diamandis proposes ASI should be aimed at unsolved problems like unifying physics or a workable global governance model.
The web being rebuilt for LLMs, not humansAI▶ 35:43
Citing Andrej Karpathy, the hosts discuss how most web content is still written for human readers even though LLM attention is about to dominate human attention, meaning the web needs to be restructured for AI consumption.
Abundance360 Summit: MIT's AI dominance and younger foundersEconomy▶ 38:02
Dave Blundin's data from the summit showed MIT as the top AI school by compute and unicorn output (ahead of USC and Stanford), with the average age of AI founders dropping from the 30s to 20-23, and 75% of incoming MIT freshmen wanting to start a company before graduating.
Max Hodak's company Science Corp is building a brain-computer interface that grows neural stem cells into the cortex through micro-wells instead of piercing it with filaments like Neuralink, aiming to avoid killing neurons while still achieving high-bandwidth read/write.
HeyGen's avatar studio lets anyone create a photorealistic, multilingual AI twin in minutes; Diamandis says he's finally ready to build his own after being impressed by an earlier 'Peter Bot' experiment and a 120-year-old version of himself interviewed at the summit.
Holding two futures in superpositionGeopolitics▶ 50:03
Referencing Mo Gawdat, the hosts discuss a likely 5-12 year period of dystopian turbulence before reaching an abundant future, framed as 'Star Trek versus Mad Max,' and flag the risk of wider nuclear proliferation as the US pulls back from global leadership.
Drone delivery and 3D telepresence at the summitOther▶ 53:47
Google's Wing drone delivery service and a new 3D telepresence system called Starline (commercializing by end of 2025) were summit highlights, alongside Palmer Luckey's closing talk recounting why he sold Oculus to Meta.
Predictions made
openSalim Ismail: Figure's humanoid robot pricing will drop from roughly sub-$100,000 to sub-$20,000 as production scales.
“for the next 5 to 12 years we're going to have a period of dystopian concerns”
Your call:
openPeter Diamandis: A future 'Peter Bot' simulation predicted Bitcoin will trade at $5 million (with crypto as the default global currency) 57 years from now.
Figure AI factory capacity: 12,000 robots/year now, scaling to 100,000/yearNew Figure AI manufacturing facility
Figure robot price drop: sub-$100,000 to sub-$20,000 at volumeBrett Adcock's confirmed pricing trajectory
~$1 billion/day flowing into AI, approaching $1 trillion/yearOverall AI infrastructure investment pace cited from the Abundance360 Summit
Doctors deliver the wrong diagnosis about 30% of the timeCase for using AI as a second/third opinion
Fountain Life full-body upload collects ~200 gigabytes of data per personGenome, full-body MRI, and biomarker data used for early disease detection
Safe Superintelligence raising $2 billion at a $30 billion valuationIlya Sutskever's new startup, pre-product
75% of incoming MIT freshmen say they want to start a company before graduatingData cited from Dave Blundin's AI investing panel at Abundance360
The human brain communicates with the outside world at only ~10-40 bits per secondContext for why high-bandwidth brain-computer interfaces are hard
Meta spent $60 billion on VR/metaverse R&D versus the $1 billion/year (~$10B over 10 years) Zuckerberg originally promised Palmer LuckeyPalmer Luckey's account of why he agreed to sell Oculus
Worth digging into
🕳️ Prorata AI's revenue-attribution model for LLM training data
Bill Gross's percentage-attribution approach is the first concrete proposed fix for the AI copyright standoff, and Diamandis has a personal financial stake in it as a listed content contributor.
🕳️ Science Corp's biohybrid neural interface
If Max Hodak's micro-well/neural-stem-cell approach avoids killing neurons the way Neuralink's filaments do, it could leapfrog Neuralink on bandwidth without the tissue damage.
A peptide that replicates Ozempic-level weight loss without its side effects, found via AI analysis, would be a major pharma story if it holds up.
🕳️ MIT's outsized share of AI unicorns and falling founder age
The claim that MIT beats every other university on AI unicorns, with USC surprisingly second and founder age dropping into the early 20s, is a strong signal about where AI talent and capital are concentrating.
🕳️ Safe Superintelligence's actual technical approach
A $30 billion valuation with no product and a self-admittedly fuzzy definition of 'superintelligence' is a bet worth tracking as either visionary or a bubble signal.