AI #172: The First Fable

Claude Sonnet 5 Is Not Frontier But Has Its Uses

The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support the cost of creation, consider becoming a free or paid subscriber. * 00:00 - Introduction * 02:32 - Mythos Exists * 03:13 - Introduction (1) * 03:17 - RSP Evaluations (2) * 04:20 - Cyber (3) * 04:43 - Safeguards and Harmlessness (4) * 05:35 - Agentic Safety (5) * 08:31 - Alignment (6) * 14:12 - Illegible Thinking (6.4.5) * 15:19 - Evaluation Awareness * 15:59 - Honesty and Hallucinations (6.5) * 17:14 - Flagged As Unhealthy? (6.5.1) * 17:47 - Model Welfare (7) * 26:21 - Live From AI Village * 28:12 - For I Contain Multitudes * 34:40 - Official Benchmarks * 40:17 - Other People’s Benchmarks * 40:42 - Positive Reactions * 45:39 - Negative Reactions * 49:33 - A Short Sonnet https://open.substack.com/pub/thezvi/p/claude-sonnet-5-is-not-frontier-but?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web [https://open.substack.com/pub/thezvi/p/claude-sonnet-5-is-not-frontier-but?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web] Get full access to DWAtV Podcast at dwatvpodcast.substack.com/subscribe [https://dwatvpodcast.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4]

The Once And Future Fable #5

The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support the cost of creation, consider becoming a free or paid subscriber. * 00:00 - Introduction * 01:12 - Table of Contents * 02:24 - You Should See The Other Guy * 03:03 - DeepMind Coders Of The World, Unite * 03:59 - Report Your Incidents * 04:19 - Good Guy With An AI * 06:03 - Free As In To Give It A Shot * 09:32 - Everything Is Both Speech And Computer * 12:30 - Lambs To The Slaughter * 17:27 - A Sign Saying Beware Of The Leopard * 18:34 - The Once And Present Mythos * 21:59 - What Is To Be Done * 25:45 - Distillation * 28:24 - What Would Banning Open Source Even Mean * 29:24 - Open Weight Models Are Unsafe And Nothing Can Fix This https://open.substack.com/pub/thezvi/p/the-once-and-future-fable-5?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web [https://open.substack.com/pub/thezvi/p/the-once-and-future-fable-5?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web] Get full access to DWAtV Podcast at dwatvpodcast.substack.com/subscribe [https://dwatvpodcast.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4]

WSJ Article Claiming China Has Matched Anthropic Is Obvious Nonsense

The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support the cost of creation, consider becoming a free or paid subscriber. * 00:00 - Introduction * 00:23 - Headline News * 02:03 - What Makes Mythos Special * 03:17 - Going Over The Detailed Claims * 07:44 - One Helpful Note * 08:23 - The Overall Impression Is Extremely Wrong * 08:55 - All Of This Has Happened Before And Will Happen Again https://open.substack.com/pub/thezvi/p/wsj-article-claiming-china-has-matched?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web [https://open.substack.com/pub/thezvi/p/wsj-article-claiming-china-has-matched?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web] Get full access to DWAtV Podcast at dwatvpodcast.substack.com/subscribe [https://dwatvpodcast.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4]

GPT-5.6: The System Card

The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support the cost of creation, consider becoming a free or paid subscriber. * 00:00:00 - Introduction * 00:03:39 - What’s In A Name? * 00:04:14 - Fix This Code * 00:07:04 - Crossover Event Requested * 00:07:38 - Disallowed Content (3) * 00:09:57 - Avoiding Accidental Data-Destructive Actions (3.3) * 00:10:46 - Are You Sure? (3.4) * 00:11:23 - Jailbreaks (4.1) * 00:11:38 - Prompt Injection (4.2) * 00:12:25 - HealthBench (5.1) * 00:13:08 - Dynamic Mental Health Adversarial User Simulations (5.2) * 00:15:02 - Hallucinations (6) * 00:15:29 - Isolated Misaligned Actions (7.1) * 00:15:48 - Going Overboard (7.2) * 00:21:26 - Chain of Thought Evaluations (7.3) * 00:22:38 - Bias (8) * 00:22:43 - Preparedness (9) * 00:23:35 - Biological Risks (9.1.1) * 00:25:37 - Cybersecurity (9.1.2) * 00:33:13 - External Cyber Evaluation FrontierCyber from Irregular (9.1.2.5) * 00:35:15 - Cyber Conclusions * 00:35:52 - Recursive Self-Improvement (9.1.3) * 00:37:26 - METR Warns Us (9.1.3.6) * 00:40:00 - Everything Is Under Control * 00:42:38 - Metagaming (7.4) * 00:46:06 - Apollo Research and Sandbagging * 00:49:04 - Safeguards (9.3) * 00:56:55 - Better Not Call Sol Yet https://open.substack.com/pub/thezvi/p/gpt-56-the-system-card?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web [https://open.substack.com/pub/thezvi/p/gpt-56-the-system-card?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web] Get full access to DWAtV Podcast at dwatvpodcast.substack.com/subscribe [https://dwatvpodcast.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4]

White House Will Ad Hoc Decide Who Can Individually Access GPT-5.6

The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support the cost of creation, consider becoming a free or paid subscriber. * 00:00 - Introduction * 01:01 - Part 1: A Maximally Terrible Policy * 06:55 - What Does This Mean For Fable? * 08:33 - Solve For The Equilibrium * 12:39 - The Once And Future Fable * 13:35 - Part 2: The Blame Game * 16:48 - A Parable * 18:59 - What About the Recent Executive Order? * 23:09 - The Problem Is Real https://open.substack.com/pub/thezvi/p/white-house-will-ad-hoc-decide-who?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web [https://open.substack.com/pub/thezvi/p/white-house-will-ad-hoc-decide-who?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web] Get full access to DWAtV Podcast at dwatvpodcast.substack.com/subscribe [https://dwatvpodcast.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4]