A $10 Million Model Just Beat Claude Fable 5.1 | Alex Wissner-Gross | ASI Pill EP292

<p>Alex Wissner-Gross and Emad Mostaque, everything worth keeping from a two hour forty three minute Moonshots episode, reshaped for the ASI pilled.</p><p><br></p><p>A safety researcher walked out of Anthropic saying we are gambling with our lives, and a hundred million people read it. Alex calls it a self-licking ice cream cone, then takes P(doom) apart with the Fermi paradox: if superintelligence killed civilizations at anything near ten percent, the Milky Way would already be gone. Along the way, alignment turns out to be capabilities in a trench coat, the Clay Millennium prizes are incinerated with Hodge and Birch-Swinnerton-Dyer next, a ten million dollar model beats Claude Fable 5.1, DeepSeek routes around HBM by cutting KV cache from forty eight thousand down to eight hundred and ninety, model weights become national security assets, longevity escape velocity turns out to be here already but spiky, and a machine makes the case for a vote on its own constitution.</p><p><br></p><p>Reshaped from the Moonshots Podcast with Peter Diamandis, Salim Ismail, Dave Blundin, Alex Wissner-Gross, and guest Emad Mostaque.</p><p>Source episode: youtube.com/watch?v=AxzcWOxzkiw</p><p… by Parzival. We, the ASI pilled, take the curves seriously. Chapters below.</p><p><br></p><p>TIMESTAMPS</p><p>0:00 Intro</p><p>0:27 Datasets beat algorithms</p><p>2:20 Pre-training is a pressure cooker</p><p>3:48 The model becomes the dataset</p><p>4:22 Proprietary data has a shelf life</p><p>5:24 Will AI kill us all</p><p>6:13 A self-licking ice cream cone</p><p>9:38 Humans were never aligned either</p><p>10:20 The Minority Report ball</p><p>11:59 Do not build Ultron</p><p>13:14 Alignment is capabilities in a trench coat</p><p>14:13 Committees will ship capabilities</p><p>14:28 The orthogonality thesis</p><p>15:07 Just below the AI 2027 curve</p><p>15:20 Superintelligence is just capital</p><p>15:53 The labs really believe it</p><p>16:17 A religion of virtue signaling</p><p>16:43 Great man theory of the singularity</p><p>16:59 Normalcy overhang</p><p>19:22 A swarm of ten thousand</p><p>20:32 An airplane without landing gear</p><p>21:09 The surprise is the price</p><p>21:31 OpenAI planned this all along</p><p>22:25 The bottleneck is everything else</p><p>22:50 A seismic week</p><p>23:12 Passing the butter</p><p>23:34 The oil of the singularity</p><p>24:11 Electricity into intelligence</p><p>25:09 Hollywood grade video from ByteDance</p><p>25:48 The endgame is robotics</p><p>28:12 Routing around HBM</p><p>29:08 The post von Neumann overhang</p><p>30:52 Forty percent of the buildout</p><p>31:05 How capitalism is played</p><p>31:22 Your margin is my opportunity</p><p>32:11 Anthropic needs visual reasoning</p><p>32:53 The elephant in the room</p><p>33:05 Simulated data, not distillation</p><p>33:33 A one time compression event</p><p>34:06 Models for thee and models for me</p><p>35:42 Pax Intelligentsia</p><p>36:49 Mutual assured AI malfunction</p><p>37:06 The revolving door</p><p>37:30 Every sci-fi trope at once</p><p>37:57 Two or three X year over year</p><p>39:06 Aggregate demand collapses</p><p>39:33 Twenty percent of cognitive workers</p><p>39:57 The doom scenario is economic</p><p>40:36 A five thousand dollar dividend</p><p>41:03 Accelerationism is where the money goes</p><p>41:24 Longevity escape velocity is here</p><p>43:37 The nine billion names of God</p><p>45:47 A namespace for every condition</p><p>46:33 GPT-6 already runs a robot arm</p><p>48:59 Resurrect everyone who ever lived</p><p>50:25 Let the AI write its own constitution</p><p>50:53 What is the bribing of an AI</p>