Bitcoin’s Bull Run & the AI Arms Race: What You Need to Know w/ Salim Ismail | EP #166 [REUPLOAD]
Peter Diamandis and Salim Ismail open with China's all-in robotics push versus the West's slower rollout, then dig into frontier-model IQ benchmarks and why Gemini 2.5 leads on performance while OpenAI dominates on revenue thanks to ChatGPT's 'user interface moment.' A clip of Demis Hassabis and commentary from Dario Amodei anchor a segment on AI-accelerated medicine and longevity, including Peter's own diagnostic, microbiome, and skincare stack. The pair cover Anthropic's values-in-the-wild study, Thinking Machines Lab's record $2B seed round, Google's shift toward real-world experiential training data, and the AI 2027 forecasting paper's US-vs-China bioweapon and abundance scenarios. They close on Bitcoin's push back above $90,000 and Peter's long-held binary, HODL-and-hold thesis.
Peter contrasts a Beijing airport robot showcase with fashion ads at JFK, and Salim argues China is racing to automate its workforce ahead of a demographic crunch.
Peter shows a chart plotting frontier models (o3 ~133, Gemini 2.5 ~127, Claude 3 at 101 eighteen months ago) against the human IQ bell curve, debating what IQ even captures once models exceed Mensa-genius level.
Humanity's Last Exam and the AI leaderboardAI▶ 7:47
Discussion of an AI benchmark testing reasoning across obscure academic fields, with Gemini 2.5 leading most performance metrics while OpenAI dominates on revenue.
OpenAI's revenue lead and the user-interface momentAI▶ 11:06
Peter and Salim unpack why OpenAI monetizes better despite Google's technical lead, crediting ChatGPT as a 'user interface moment' akin to Mosaic, the iPhone, and Coinbase that made a complex technology instantly usable.
AI, DeepMind, and the end of diseaseLongevity▶ 17:10
A 60 Minutes clip of Demis Hassabis describing AI-driven drug discovery timelines shrinking from years to weeks, and the pair's take on wearable-driven, real-time AI diagnosis.
Peter details three products he personally uses to extend healthspan: Fountain's full-body diagnostic upload, Viome's microbiome metatranscriptomics testing, and the OneSkin peptide for senescent skin cells.
A study of 700,000 anonymized Claude conversations surfaced five broad value categories (helpfulness, accuracy, empathy, safety, authenticity), which Peter and Salim read as evidence AI can carry a consistent moral code.
Constitutional alignment and geopolitical rivalryAI▶ 30:00
Salim floats training AI on the US Constitution plus UN human rights documents to solve alignment, while Peter raises the open question of what founding documents China or Russia will train their own AI on.
Thinking Machines Lab's record seed roundEconomy▶ 30:49
Mira Murati's Thinking Machines Lab raised $2B at a $10B valuation, double what she originally sought two months earlier, prompting a debate on fundraising discipline versus froth.
Training AI on real-world and deep-web dataAI▶ 34:38
A Google paper argues AI is outgrowing human-made data; Peter and Salim discuss Google and xAI's unique access to street view, YouTube, Gmail, X, and Tesla data, plus the untapped deep web.
Peter walks through a forecasting paper imagining a US ('OpenBrain') vs China ('DeepCent') AI race that splits into a 'go fast' path ending in a bioweapon catastrophe and a 'slow down' path ending in aligned abundance.
With Bitcoin back above $90,000, Peter restates his binary thesis (zero or past $1M) and his 'HODL as forced savings' approach, while Salim notes recent growth spikes rewarded only those already holding.
Predictions made
openSalim Ismail: China will keep mass-deploying robots to automate its workforce because its population decline will leave too few workers within the next decade or two.
“we're getting ready for a monster bull run in Bitcoin”
Your call:
Numbers that matter
Frontier AI models now score above average human IQ: OpenAI o3 ~133, Gemini 2.5 ~127, GPT-o1 ~120, Claude 3 reached 101 about 18 months earlierPlotted against the human IQ bell curve (average 100, genius/Mensa ~140) to argue AI is already past genius-level on this measure.
As of end of December 2024: OpenAI ~$2.5B revenue vs Gemini just under $0.5B, with Anthropic lower stillCited as evidence OpenAI is winning on revenue despite Gemini's performance lead, and doesn't yet reflect OpenAI's gains in the following four months.
AI-era health monitoring has grown from 4 basic vitals to roughly 40 live wearable data streamsSalim contrasts old heart-rate/blood-pressure/glucose tracking with today's continuous biometric streams AI can correlate against disease.
Fountain's full-body diagnostic upload captures about 200GB of personal dataPeter uses this annual scan to catch cardiovascular, cancer, neurodegenerative, and metabolic disease at inception.
Thinking Machines Lab raised $2B at a $10B valuation, roughly double what it was seeking two months earlierCalled the largest seed round in history.
Roughly $1 billion per day is being invested into the AI spaceCited from Peter's Abundance Summit discussions as context for the mega seed-round froth.
Bitcoin trading back above $90,000Starting point for the episode's bull-run discussion and Peter's 'all in' thesis.
Bitcoin's two biggest single-day gains in 2024 were an $8,000 jump on Nov 12 and a near-10% jump on Feb 28Salim and Peter note that missing those exact days meant missing most of the year's growth.
Worth digging into
🕳️ The AI 2027 scenario paper (OpenBrain vs DeepCent)
A detailed, choose-your-own-adventure forecast from AI policy experts mapping both a bioweapon-catastrophe path and an aligned-abundance path by 2030 — a concrete model for what 'fast takeoff' could actually look like.
🕳️ Anthropic's 700,000-conversation values study
First large-scale empirical look at what values a deployed LLM actually exhibits in the wild, not just what it's trained to say.
🕳️ Colossal's de-extinction tooling and embedded ethics teams
The dire-wolf project is a minor headline next to the real story: a generalizable DNA-editing tool set, paired with a standing ethics review on every project, that's directly transferable to human disease editing.
🕳️ Viome's metatranscriptomics, born from Los Alamos biodefense research
An unusual technology lineage, from nuclear-bomb-era biodefense science to personalized nutrition, that's worth understanding before recommending the product.