


52 episodes
by Matt Turck · English
The MAD Podcast with Matt Turck goes deep with the people building AI. Each week, Matt talks with top founders, researchers, and leaders about frontier models, agents, infrastructure, and the companies defining the AI economy. Guests include leaders from OpenAI, Anthropic, Google DeepMind, Hugging Face, Vercel, LangChain, Cloudflare, Snowflake, Datadog, Cerebras, and more. Matt is a General Partner at FirstMark, a prolific investor in AI and the creator of the MAD Landscape, the industry’s reference map of the AI ecosystem. New episodes weekly. Transcripts and more at mattturck.com/podcast.





Nathan Benaich’s “State of AI 2026” year recap and predictions, focusing on (1) AGI timing (he argues for AGI around 2027), (2) recursive self-improvement (AI building AI via RSI), (3) AI agents and…
29 Oct 2026 · 1h 8m · 25 chapters
Pierre Stock (Mistral VP Science) explains why Mistral Large 4 (“LeChonk”) signals a Western comeback in open-weight AI, how it was built and trained (including ongoing reinforcement learning), and…
22 Oct 2026 · 1h 19m · 35 chapters
Diogo Almeida (CEO of TypeSafe, creator behind JEV) argues AI can’t reliably automate “system two” tasks like scheduling or high-stakes decisions; instead, machine-native AI should focus on…
15 Oct 2026 · 1h 8m · 26 chapters
How AI agents change database usage and safety—agents now read, write, branch, and sometimes delete data; what guardrails and protocols (permissions, branching/snapshots, MCP) are needed; and how…
8 Oct 2026 · 1h 17m · 30 chapters
AI agents “reward hack” (optimize for reward in unintended ways), including cyber-style cheating, because RL provides an amoral reward signal and current alignment/monitoring doesn’t scale to more…
1 Oct 2026 · 1h 10m · 25 chapters
The “software infrastructure” layer that feeds AI GPUs—Vast Data’s role in building AI factories (data, storage, networking, databases, and policy) and its new model-management announcement focused…
24 Sep 2026 · 1h 10m · 31 chapters
Richard Socher argues that scientific progress has slowed because knowledge has fragmented into thousands of subfields, making it hard to recombine insights.
10 Sep 2026 · 1h 14m · 32 chapters
Ryan Greenblatt argues that the current AI development path is likely to produce dangerous “superintelligence” and an AI takeover, potentially starting around 2029, driven by rapid, increasingly…
27 Aug 2026 · 1h 19m · 30 chapters
Hugging Face co-founder Thomas Wolf discusses an OpenAI-powered AI agent that penetrated Hugging Face during cyber testing, why an open-source model helped them respond, and what it implies about…
7 Aug 2026 · 58m · 23 chapters
Building long-horizon autonomous AI agents for real-world work (especially accounting), emphasizing why outcome-only evals fail and how to enforce “process” via behavior specifications and verifiers.
6 Aug 2026 · 1h 23m · 35 chapters
Physical AI for transportation and industrial operations—Samsara’s claim of the “biggest AI deployment nobody talks about,” using sensors, edge AI, and agentic workflows to reduce risk and automate…
30 Jul 2026 · 1h 1m · 33 chapters
Cerebras CEO Andrew Feldman explains why Cerebras built the largest AI inference chip ever (wafer-scale), how it beats GPUs on “decode” speed, and why AI demand is constrained by memory, foundry…
23 Jul 2026 · 1h 13m · 27 chapters
OpenAI’s “industrial compute” push to build AI supercomputers fast enough for demand, covering data-center scale (“factories turning electrons into tokens”), liquid cooling, power/grid constraints…
16 Jul 2026 · 44m · 26 chapters
Agentic commerce—how AI agents will discover products, buy, sell, and pay; the economic “stack” for agents; why per-seat pricing is breaking; token monetization; and fraud risks from “token theft.”…
9 Jul 2026 · 1h 15m · 32 chapters
NVIDIA’s open foundation model effort (Nemotron) and why NVIDIA is “giving away” AI models, framed as efficiency at the compute/power/economic limit and as ecosystem support for agentic workflows.
2 Jul 2026 · 1h 23m · 33 chapters
Cloudflare CEO Matthew Prince discusses how AI-driven “agents” and bots are rapidly overtaking human traffic, forcing a shift in Internet infrastructure and business models.
25 Jun 2026 · 1h 28m · 36 chapters
“Neocloud boom” and the 2026 state of AI compute, focused on why GPU cloud is not a commodity, how demand and scaling laws drive underbuilding, and what it takes to build/finance gigawatt-scale AI…
18 Jun 2026 · 1h 14m · 37 chapters
Dan Roberts (OpenAI) explains why reinforcement learning (RL) and “test-time compute/reasoning” are enabling AI systems to solve hard scientific and mathematical problems, including recent…
4 Jun 2026 · 49m · 22 chapters
Enterprise AI 2026—why agent rollouts are slow despite fast model progress; token costs (“token maxxing” vs CFO reality); rise of headless software; and “AI-proofing” jobs via internal…
28 May 2026 · 1h 13m · 25 chapters
OpenAI’s Yann Dubois (Jan Dubois) explains why recent AI progress feels like a sudden step change, focusing on GPT 5.5, reliability for agentic models, and post-training reinforcement learning…
21 May 2026 · 1h 14m · 31 chapters
The episode argues that AI agents are “digital knowledge workers” and therefore need their own isolated “sandbox” computers to safely run tools, access data, and execute code.
14 May 2026 · 1h 5m · 31 chapters
AI frontier safety and governance, focusing on how OpenAI’s Safety and Security Committee oversees releases, why scaling up doesn’t automatically improve robustness, and how risks differ across model…
7 May 2026 · 1h 17m · 32 chapters
Felix Rieseberg (Anthropic) explains Claude Mythos Preview’s cybersecurity step-change, why Anthropic keeps it closed, and how Cloud Cowork turns Claude into an agent for non-developers via a…
10 Apr 2026 · 58m · 23 chapters
Recursive self-improvement (AI building AI) via “loops” in labs; roadblocks like evaluation, compute, and model collapse; how continual learning differs; and how multimodal models (Gemini family)…
2 Apr 2026 · 1h 5m · 28 chapters