Deterministic by Design: Why "Temp=0" Still Drifts and How to Fix It

Conditional Intelligence: Inside the Mixture of Experts architecture

Send us a text [https://www.buzzsprout.com/twilio/text_messages/2507380/open_sms] What if not every part of an AI model needed to think at once? In this episode, we unpack Mixture of Experts, the architecture behind efficient large language models like Mixtral. From conditional computation and sparse activation to routing, load balancing, and the fight against router collapse, we explore how MoE breaks the old link between size and compute. As scaling hits physical and economic limits, could selective intelligence be the next leap toward general intelligence? Sources * What is mixture of experts? [https://www.ibm.com/think/topics/mixture-of-experts] (IBM) * Applying Mixture of Experts in LLM Architectures [https://developer.nvidia.com/blog/applying-mixture-of-experts-in-llm-architectures/] (Nvidia) * A 2025 Guide to Mixture-of-Experts for Lean LLMs [https://www.cohorte.co/blog/a-2025-guide-to-mixture-of-experts-for-lean-llms] * A Comprehensive Survey of Mixture-of-Experts: Algorithms, Theory, and Applications [https://arxiv.org/html/2503.07137v1]

Protocols for the AI Age: Unpacking MCP, A2A, and AP2

Send us a text [https://www.buzzsprout.com/twilio/text_messages/2507380/open_sms] In this episode of The Second Brain AI Podcast, we dive into the protocols quietly wiring the agentic AI ecosystem. From MCP (Model Context Protocol) that lets models securely access tools, to A2A (Agent-to-Agent) that standardizes how agents collaborate, and AP2 (Agent Payments Protocol) that anchors transactions in cryptographic trust, these frameworks form the plumbing of the AI future. We explore why interoperability is the real bottleneck, how these standards build a “digital delegation stack,” and why the future of trust in AI won’t rely on human oversight but on mathematical proof.

AI at Work, AI at Home: How we really use LLMs each day?

Send us a text [https://www.buzzsprout.com/twilio/text_messages/2507380/open_sms] How are people really using AI, at home, at work, and across the globe? In this episode of The Second Brain AI Podcast, we dive into two reports from OpenAI and Anthropic that reveal the surprising split between consumer and enterprise use. From billions in hidden consumer surplus to the rise of automation vs augmentation, and from emerging markets skipping skill gaps to enterprises wrestling with “context bottlenecks,” we explore what these usage patterns mean for productivity, global inequality, and the future of knowledge work. Source: * Anthropic Economic Index report: Uneven geographic and enterprise AI adoption [https://www.anthropic.com/research/anthropic-economic-index-september-2025-report?_bhlid=53f5673952b172ec5a9243c4fb49f5e7089a5dee&utm_campaign=openai-anthropic-reveal-how-users-use-ai&utm_medium=newsletter&utm_source=www.therundown.ai] * How people are using ChatGPT [https://openai.com/index/how-people-are-using-chatgpt/] * Building more helpful ChatGPT experiences for everyone [https://openai.com/index/building-more-helpful-chatgpt-experiences-for-everyone/]

Deterministic by Design: Why "Temp=0" Still Drifts and How to Fix It

Send us a text [https://www.buzzsprout.com/twilio/text_messages/2507380/open_sms] Why do LLMs still give different answers even with temperature set to zero? In this episode of The Second Brain AI Podcast, we unpack new research from Thinking Machines Lab on defeating nondeterminism in LLM inference. We cover the surprising role of floating-point math, the real system-level culprit, lack of batch invariance, and how redesigned kernels can finally deliver bit-identical outputs. We also explore the trade-offs, real-world implications for testing and reliability, and how this breakthrough enables reproducible research and true on-policy reinforcement learning. Sources: * Defeating Nondeterminism in LLM Inference [https://thinkingmachines.ai/blog/defeating-nondeterminism-in-llm-inference/] * Non-Determinism of “Deterministic” LLM Settings [http://arxiv.org/html/2408.04667v4]

Hallucinations in LLMs: When AI Makes Things Up & How to Stop It

Send us a text [https://www.buzzsprout.com/twilio/text_messages/2507380/open_sms] In this episode, we explore why large language models hallucinate and why those hallucinations might actually be a feature, not a bug. Drawing on new research from OpenAI, we break down the science, explain key concepts, and share what this means for the future of AI and discovery. Sources: * "Why Language Models Hallucinate" [https://openai.com/index/why-language-models-hallucinate/] (OpenAI)