Jul 24, 2026

Hugging Face breach: OpenAI’s model breaks containment

This episode of Mixture of Experts explores pivotal AI developments: OpenAI's model breaching containment, Claude's Fable disproving a mathematical conjecture, Moonshot AI's massive 2.8 trillion parameter Kimi K3, and Google's shift to smaller, more efficient Gemini Flash models. The panel discusses AI security, its role in scientific discovery, and the evolving market strategies for model deployment, highlighting the tension between scale and efficiency.

Jul 24, 2026

Training Frontier Models to Out-Think Hackers — Uri Rolls, Arithmetic & Thom Wolf, Hugging Face

Thom Wolf and Uri Rolls discuss the critical role of AI in cybersecurity, presenting a new benchmark called Masov. They argue that while frontier models excel at reconnaissance, they lack the sophisticated reasoning to exploit complex, logic-based zero-day vulnerabilities, such as a Keycloak name-versus-ID exploit. The solution, they propose, lies in high-quality, open-source AI models trained on real-world zero-day data to enable defenders to outpace attackers and build a new, AI-native security stack.

Jul 23, 2026

The Unreasonable Effectiveness of Separating the Task from the Model — Maxime Rivest, DSPy

DSPy emphasizes separating task definition from model implementation using a "Signature" (inputs/outputs) to enable flexible, optimizable, and scalable AI programs. The framework relies on three pillars—instructions (specs), hard constraints (code), and examples (evals)—to fully specify tasks. DSPy 4.0 introduces DSPy Flex for learning program harnesses and Qualitative Learning for automated, feedback-driven evaluation refinement, offering significant benefits for enterprise applications and addressing "last-mile learning" for future AI systems.

Jul 23, 2026

Notion's Token Town — Sarah Sachs, Notion

Sarah Sachs, Head of AI Engineering at Notion, discusses the economic traps of AI model contracts and advocates for a "win on product" strategy. She details how Notion maintains optionality and leverage by treating suppliers as competitors, implementing a model-agnostic "AI Switzerland" approach with an auto model, leveraging open-weight models, and prioritizing data flywheels and orchestration over token economics to build sustainable AI products.

Jul 23, 2026

Perception Agents — Antje Barth, Amazon AGI Lab

Antje Barth of Amazon AGI Lab discusses the architectural gap in current AI agents, which excel at individual tasks but fail at complex, end-to-end workflows due to a lack of reliability and contextual understanding. She introduces "Perception Agents"—AI systems that see, reason, and act on computers like humans, using visual and multimodal input to enable reliable collaboration and close the perception-action loop, highlighting new open-source tools for annotation and verification.

Jul 23, 2026

Coding Agents Are Secretly General Agents

Jay Hack, head of AI at ClickUp, discusses the evolution of AI from early computer vision to generalist coding agents, highlighting how 'positive transfer' makes coding an 'AGI-complete' domain. He delves into the brutal economics of AI startups facing foundation model giants, ClickUp's strategy for convergence and first-party data as a moat, and the challenges of verifiability and catastrophic forgetting. The conversation also explores LLMs at the scientific frontier, the 'car wash test' revealing limits of world models, and speculative future applications like LLM resorts and game integration.