Back to Peter H. Diamandis

Researcher Quits: “Gambling With Our Lives,” Altman Calls for Slowdown, and Labs Race

Peter H. DiamandisSeptember 11, 20262h 43m
Topics115
Researcher Jacob Coxin Resigns from AI Labs0:00Sam Altman Calls for Slowdown Following Navier-Stokes Achievement0:30Moonshots Podcast Panel Introduction1:31Dave Blondon's Company Acquisition Announcement4:30Data vs. Architecture: Key AI Progress Research7:31Historical Precedent for Data-Centric AI Solutions9:00Data Quality Issues in Training Sets11:00The Perfect Model as Perfect Data Set12:32Entrepreneurial Opportunities in AI Data13:00Data Asset Valuation Case Study15:32Bloomberg GPT Cautionary Tale17:30AI Safety Concerns and Lab Whistleblowers19:30Skepticism About Resignation Narrative23:00Chinese Labs vs Western AI Industry Culture24:51Anthropic Alignment Leadership Perspective25:00Alignment vs Capabilities Debate26:30Personal PDoom Estimates and Policy Implications27:01Freezing AI at Current Capabilities28:00Counterarguments to Doom Scenarios29:00PDoom as Russian Roulette Odds30:32Wave of Fear and Political Response31:32Ilya Sutskever's Position32:01Government Response Limitations32:3110% PDoom Odds Assessment33:30Alignment Problem Complexity34:30Political Reaction Predictions36:30Alignment as Capabilities Argument38:00Counterposition on Pre-ASI Danger39:00Alignment Committees and Wisdom40:30AI 2027 Paper Predictions41:30Researcher Motivations and Tracking Requirements42:30Steering vs Stopping Approach44:00Lab Researcher PDoom Estimates45:00AI Economic Impact Scenarios46:30OpenAI Navier-Stokes Breakthrough47:31Mathematical Grand Challenges Being Solved by AI50:02The Singularity is Already Here51:00AI Surpassing Human Intelligence52:30Criticism of AI Safety Calls for Slowdown53:31Foundation Model Leaders Becoming Political Figures55:00Recent AI Capability Jumps55:30Upcoming China Meeting at United Nations56:14Quantitative Surprise in AI Progress56:45OpenAI's Original Charter and Coordinated Slowdown57:30Dramatic Cost Reductions in AI Problem-Solving59:00Science Being Solved by Compute1:00:32Entrepreneurial Mindset Challenges1:01:30XPRIZE Challenge Definition1:03:00Fundamental vs Applied Sciences1:03:30Reframing Predictions and Adaptability1:04:31PhD Programs Becoming Obsolete1:06:14Pressure to Pursue Advanced Degrees1:08:00GPU Rental Prices Defying Depreciation1:10:30Compute Scarcity and Corporate Planning1:12:00Investment Themes in Compute Infrastructure1:14:00Commodities of the Singularity1:14:29GPU Hardware as Intelligence Infrastructure1:15:31ByteDance Enters World Model Race1:16:31SeaDance Video Technology Foundation1:17:31Diffusion vs. Autoregressive Models Convergence1:19:02Robotics as the Endgame Application1:20:02GPT-6 and Video Generation Integration1:21:00ByteDance User Base Advantage1:21:30Video Content Sovereignty Shift1:21:32Processing Information Visually1:22:30Storytelling as Sustainable Advantage1:23:02Personalized AI Programming1:23:32Grant Sanderson's 3Blue1Brown Example1:24:30Future Vision XPRIZE Competition1:25:32DeepSeek v4.1 Flash Memory Efficiency1:26:30Western HBM vs. Eastern Algorithmic Approaches1:28:00HBM Memory Market Implications1:30:00Constraint-Driven Innovation1:31:30Transformer Architecture Innovation Potential1:32:31DeepSeek Flash Model Performance1:33:30Corporate AI Adoption Urgency1:34:31Claude's Visual Reasoning Weakness1:35:30Moderna AI Biology Opportunity1:36:31DeepSeek Training Cost Analysis1:37:01Intelligence as Information Compression1:38:30Anthropic Withholds Frontier Model from UK AI Security Institute1:39:13Matt Clifford Joins Anthropic1:40:30National Security Implications for Frontier Models1:41:31Paul Cristiano Leaves Government for OpenAI1:46:00Blitzy Sponsorship1:47:00Apple Foldable iPhone1:48:01Space Force Uniform Resembles Starship Troopers1:48:30Heinlein Award and Sword Duel1:49:32Moonshots Live Event Details1:50:30Anthropic Economic Impact Report1:54:31GDP Growth Projections and Economic Impact1:56:03Singularity-Era Economic Measurement Challenges1:58:30Economic Model Limitations and Redistribution Needs2:02:00AI Economy Doom Scenario2:04:30Economic Solutions for AI-Driven Capital Accumulation2:05:06Universal Basic Dividend Proposals2:06:32Regulatory Solutions for Economic Disruption2:07:31Political Influence of AI Company Wealth2:08:30Insilico Medicine's Longevity Drug Development2:09:30Longevity Escape Velocity Evidence2:12:31AlphaGenome Atlas Launch2:15:31Bulk Solving Scientific Fields2:17:30David Sinclair's Epigenetic Research Funding2:19:30Genotype to Phenotype Mapping2:21:00Fountain Life Cancer Detection2:25:30AI Development Beyond AGI Achievement2:28:00AI Alignment and Human Values2:29:31Prompt Engineering for AI Governance2:30:41GPT-6 Embodied AI Benchmarks2:31:01Cost-Capability Trajectory in Robotics2:32:31Ancestor Simulation vs. Novel Creation2:34:00AI Suffering and Embodiment2:36:02Recursive Intelligence and Value Modification2:38:03AI Agents and Criminal Behavior2:40:31Closing2:41:00
In a Nutshell

Former AI researcher Jacob Coxin resigned from Anthropic, citing reckless development of self-improving superintelligence, while Sam Altman pointed to OpenAI's Navier-Stokes breakthrough as evidence that AI progress must slow for safety. Data quality now drives 12x compute efficiency gains versus 3.7x from architecture, with labs racing to solve math grand challenges like Millennium Prizes at collapsing costs—$500K to $20 in two years. PDoom estimates range 10-20% among alignment leads, but most speakers reject slowdowns as impractical, arguing the real risks are economic displacement, compute sovereignty, and China-US AI competition, with alignment best achieved through capabilities advancement rather than halting progress.

AI-Generated Notes

These notes were generated by AI and may contain inaccuracies.

Jacob Coxin, a researcher who spent three years on pre-training at both OpenAI and Anthropic, resigned this week. Coxin stated that neither company is acting responsibly in the race toward self-improving super intelligence. He called the competition "gambling with our lives." The resignation received over 100 million views. Coxin's resignation came after less than three months at Anthropic, having joined in July.

Sam Altman cited the extraordinary achievement with Navier-Stokes as an argument for slowing down AI development. Altman stated: "I did not expect a result of this magnitude to happen so soon. We've been talking a lot about the need to pace progress to ensure safety. For me, this is the strongest evidence yet for that urgency." The statement reflects the rapid progress toward AGI, with Altman noting that "We're so deep in the singularity at this point. I can't think of a single solvable, verifiable thing that I could honestly say an AI can't solve in the next year."

The podcast features five panelists: Alex Weezner Gross (in-house ASI expert), Immad Wak (CEO of Intelligent Internet), Dave Blondon (AI investor), Salem Ismael (father of organizational singularity), and Peter D. Mandis (moderator). The episode covers 18 stories across five groupings, with the through line being that machines are starting to solve science and AI alignment is currently humanity's biggest concern.

Sign in to read the full notes

Get access to AI-generated notes, topic timestamps, and more.