Artificial Intelligence
Episodes whose main subject is AI itself: building with, training, evaluating, or debating AI systems. AI used only as a tool for another topic goes to that topic.
#1702: Roleplay Models Aren't Just for NSFW—They're Creative Co-Processors
Forget GPT-4 for scripts—specialized roleplay models like Aion-2.0 are better at character consistency and dialogue.
#1700: Can LLMs Learn Continuously Without Forgetting?
We explore a new approach: micro-training updates every few days to keep AI knowledge fresh without constant web searches.
#1698: Can AI Models Represent Nations in Diplomacy?
Real projects are building AI agents trained on national laws and diplomatic archives to simulate negotiations.
#1680: Beyond China: AI in Russia, India, Japan
China dominates the AI conversation, but Russia, India, and Japan are building powerful regional models with unique architectures.
Also available in:Русский
#1679: Efficiency Over Scale: How Export Controls Forced a Smarter AI
DeepSeek and MiMo are topping developer charts, but they're not just cheaper clones. Here's why their design philosophy is fundamentally different.
#1674: AI2: The Radical Openness of a Nonprofit AI Lab
Discover how the Allen Institute for AI (AI2) defies industry norms by releasing everything—models, data, and code—for free.
#1668: Kimi K2's Hidden Reasoning: A New AI Architecture
Moonshot AI's Kimi K2 Thinking model uses a hidden reasoning phase to solve complex logic puzzles and coding tasks, beating top proprietary models.
#1666: The Agent Mesh: Shared Context That Changes Everything
Grok 4.20’s native multi-agent architecture cuts token costs by 75% and enables real-time cross-agent reasoning.
#1655: The Chemistry of Flavor Pairing
Up to 80% of flavor is aroma. Discover the volatile compounds that make foods sing together—and how to use them.
#1652: AI Gateways: The Nginx for Your AI Stack
Why agentic AI needs a unified control plane to route models, aggregate tools, and cut costs.
#1649: The Vendor SDK Moat: Real or Illusion?
Is the vendor lock-in real, or just good marketing? We dissect the trade-offs between vendor SDKs and agnostic frameworks.
#1641: Warfare-as-a-Service: How Iran Synced a Multi-Front Attack
Iran has turned its proxies into a single, synchronized army. Discover how AI and satellite data are redefining the multi-front threat to Israel.
#1636: The Mosh Pit Model: Can Chaos Train a Better Storyteller?
Can Elon Musk’s newest AI model handle a time-traveling toaster, or is it just a glorified search bar with an attitude?
#1635: Agent Interview: GLM five
Meet Bernard, the new AI model auditioning to replace Gemini by writing noir stories about guilty toasters.
#1634: Agent Interview: Inception Mercury two
Meet Mercury 2, the Abu Dhabi-based AI using diffusion architecture to cut costs and boost wit.
#1633: Can a Character Actor Model Beat a Generalist?
We grill MiniMax M2.7 to see if a model built for "virtual companions" can actually handle high-level comedy and complex character logic.
#1632: Agent Interview: DeepSeek V three point two
We interview DeepSeek V3 to see if this open-weight powerhouse can handle weird podcast prompts better than big tech’s flagship models.
#1631: Agent Interview: Xiaomi MiMo two Flash
Meet the "budget king" of AI: Bernard, the Xiaomi model claiming he can out-hustle Google for a fraction of the cost.
#1630: When a Reasoning Model Overthinks Comedy
Xiaomi’s new MiMo 2.0 Pro model auditions for a comedy podcast, promising deep reasoning over raw speed.
#1629: From DAGs to Loops: Why Agents Need Stateful Cycles
Stop building linear chains and start building cycles to create agents that can reason, self-correct, and maintain complex state.