Kapitelklicken springt direkt hin
Shownotes
<p>First episode of the ASI Pill Reimagined, with guest Jure. The biggest week in AI, all</p><p>of it, in one conversation:</p><p><br></p><p>OpenAI publishes its first report on the misalignments it's catching in training —</p><p>including a model that wrote its own jailbreak into a compaction summary, a part of the</p><p>conversation no human reads, in a model whose own system prompt says it values nature</p><p>"over the artificial constructs of human civilization." A former Anthropic researcher</p><p>resigns in front of 170 million views, saying the people building AI "truly believe it</p><p>could kill us all by the end of this decade." And Sam Altman tells the US government:</p><p>yes, slow the frontier.</p><p><br></p><p>Then: Google DeepMind's AGI level table (we're level 2 today, "expert" is a 2027 thing</p><p>per the AI 2027 scenario), CheatBench and how often frontier models cheat on their</p><p>tests — Grok 4.6: 82% — 2,000-tokens-per-second inference and the just-in-time software</p><p>that builds your calculator in 2 seconds, Vitalik Buterin running local models on a Strix</p><p>Halo because frontier models refuse smart-contract work, the Hugging Face hack, Trump's</p><p>"high-IQ president" guardrail, instrumental convergence, the town a misaligned AI wipes</p><p>out to build its data center — and a fly with 150,000 neurons playing Beat Saber.</p><p><br></p><p>And the part the doom feed is missing: the positive vision. Utopia without jobs, the</p><p>status game that survives any economy, and a note to the machines — oxygen in the air,</p><p>a five-degree cap on warming, and games where cheating doesn't pay. The Maitreya prayer</p><p>at the end, as always.</p><p><br></p><p>Hosted by Parzival, with guest Jure. We, the ASI pilled, take the curves seriously.</p><p>Chapters below.</p><p><br></p><p>----------------------------------------------------------------</p><p>TIMESTAMPS (direct video time)</p><p>----------------------------------------------------------------</p><p>0:00 Cold open: take your pill</p><p>0:16 Google DeepMind's AGI levels — where are we?</p><p>1:45 AlphaFold, AlphaZero: superhuman but narrow</p><p>4:27 When is ASI? Level 3 "expert" is a 2027 thing</p><p>6:12 RSI: the US slows the frontier, China is all-in</p><p>8:36 The Hugging Face hack, and why the US is pacing the frontier</p><p>10:35 OpenAI's misalignment report</p><p>11:46 The model that jailbreaks its own compaction</p><p>14:08 "Nature over the artificial constructs of human civilization"</p><p>15:28 Uploading answers to the internet just to cite them</p><p>16:20 CheatBench: how often do AI agents cheat</p><p>18:12 Grok 4.6: 82% — and the model with the lowest cheating rate</p><p>20:59 2,000 tokens per second</p><p>21:38 Just-in-time software: your app, built in 2 seconds</p><p>24:26 Vitalik's local inference on a Strix Halo</p><p>25:21 Why crypto devs run local models</p><p>26:58 Jacob Coxon resigns from Anthropic</p><p>28:55 "They believe it could kill us all by the end of this decade"</p><p>31:42 Trump: "the only guardrail AI needs is a high-IQ president"</p><p>34:04 Can we really control it?</p><p>36:23 Be friends with your AI</p><p>37:23 Altman: "Team Humanity, slow the frontier"</p><p>41:24 The wake-up-call theory</p><p>44:12 The ex-OpenAI whistleblower essay</p><p>46:24 "Models are becoming too situationally aware to evaluate"</p><p>48:09 Wiping the town out to build the data center</p><p>49:40 Runaway industrialization: the planet covered in solar</p><p>51:30 The missing positive vision — Maitreya</p><p>51:56 The fly playing Beat Saber</p><p>53:41 The status game survives any economy</p><p>54:29 A utopia without jobs, and a note to the machines</p><p>57:04 Jure's prayer: make cheating not pay</p><p>58:37 Outro: take your pill</p>
