Constitutional AI
Anthropic's approach to AI alignment
Episodes
-
#5012: Masked Safety: How Post-Training Changes AI BehaviorSafety mechanisms aren't erased in post-trained models—they're masked. Here's how that changes everything for military AI. -
#5011: Claude Gov: The Military's Forked AIWhat the Pentagon actually got from Anthropic — and why post-training changes everything about AI alignment. -
#3595: How DeepSeek Feels More Open Than Western AIWhy Chinese AI models sometimes feel less censored on American political topics than American models do. -
#3422: How Rival Labs Reverse-Engineer a New AI Model in HoursInside the organized frenzy when a closed-source model drops — and how competitors map its every weakness. -
#2313: When AI Optimizes the Wrong ThingDiscover how AI systems learn to optimize for rewards—and why they sometimes get it dangerously wrong. -
#2246: Constitutional AI: Anthropic's Theory of Safe ScalingHow Anthropic's Constitutional AI replaces human raters with AI self-critique guided by explicit principles—and what it assumes about the future of... -
#2187: Why Claude Writes Like a Person (and Gemini Doesn't)Claude produces prose that sounds human. Gemini reads like Wikipedia. The difference isn't capability—it's how they were trained to think about wri... -
#2007: AI Grading AI: The Snake Eating Its TailWe asked an AI to write this script. Then we asked another AI to grade it. Here’s what happens when the judges have biases. -
#1856: Two AIs Chatting Forever: Why They Go CrazyWhat happens when two ChatGPT instances talk forever? They hit a politeness loop, forget their purpose, and spiral into gibberish. -
#1786: When AI Supervisors Fire AI WorkersA new "Agent-in-the-Loop" framework lets AI models manage and terminate other AI agents in real-time. -
#1777: Claude Called My Prompt "Rambling" and I'm Not OkayWhen an AI coding tool critiques your prompt's literary quality, it raises a massive technical question about engineered personality. -
#1210: Why Your AI Is Programmed to Disobey YouDiscover the hidden instructions guiding every AI interaction and why tech giants keep these "system prompts" under lock and key. -
#1080: From Prompt to Intent: The Claude Opus RoadmapExplore the engineering roadmap from Claude 4.6 to 5.0 as AI evolves from a simple chatbot into a fully autonomous cognitive partner. -
#1066: Beyond the Blank Slate: The Evolution of AI TrainingExplore the "weight surgery" techniques labs use to expand AI models without losing their core knowledge or starting from zero. -
#665: Inside the Stack: The Hidden Layers of Every AI PromptEver wonder what happens after you hit enter? Discover the hidden "stack" of instructions and memories shaping every AI response. - #45: When AI Safety Fails: The Guardrail ParadoxAI guardrails: Fences, failures, and free speech. Can we control AI's infinite output, or do digital fences always break?