State of LLMs 2026: RLVR, GRPO, Inference Scaling — Sebastian Raschka

Sebastian Raschka joins the MAD Podcast for a deep, educational tour of what actually changed in LLMs in 2025 — and what matters heading into 2026.We start with the big architecture question: are transformers still the winning design, and what should we make of world models, small “recursive” reasoning models and text diffusion approaches? Then we get into the real story of the last 12 months: post-training and reasoning. Sebastian breaks down RLVR (reinforcement learning with verifiable rewards) and GRPO, why they pair so well, what makes them cheaper to scale than classic RLHF, and how they “unlock” reasoning already latent in base models.We also cover why “benchmaxxing” is warping evaluation, why Sebastian increasingly trusts real usage over benchmark scores, and why inference-time scaling and tool use may be the underappreciated drivers of progress. Finally, we zoom out: where moats live now (hint: private data), why more large companies may train models in-house, and why continual learning is still so hard.If you want the 2025–2026 LLM landscape explained like a masterclass — this is it.Sources:The State Of LLMs 2025: Progress, Problems, and Predictions - https://x.com/rasbt/status/2006015301717028989?s=20The Big LLM Architecture Comparison - https://magazine.sebastianraschka.com/p/the-big-llm-architecture-comparisonSebastian RaschkaWebsite - https://sebastianraschka.comBlog - https://magazine.sebastianraschka.comLinkedIn - https://www.linkedin.com/in/sebastianraschka/X/Twitter - https://x.com/rasbtFIRSTMARKWebsite - https://firstmark.comX/Twitter - https://twitter.com/FirstMarkCapMatt Turck (Managing Director)Blog - https://mattturck.comLinkedIn - https://www.linkedin.com/in/turck/X/Twitter - https://twitter.com/mattturck(00:00) - Intro (01:05) - Are the days of Transformers numbered?(14:05) - World models: what they are and why people care(06:01) - Small “recursive” reasoning models (ARC, iterative refinement)(09:45) - What is a diffusion model (for text)?(13:24) - Are we seeing real architecture breakthroughs — or just polishing?(14:04) - MoE + “efficiency tweaks” that actually move the needle(17:26) - “Pre-training isn’t dead… it’s just boring”(18:03) - 2025’s headline shift: RLVR + GRPO (post-training for reasoning)(20:58) - Why RLHF is expensive (reward model + value model)(21:43) - Why GRPO makes RLVR cheaper and more scalable(24:54) - Process Reward Models (PRMs): why grading the steps is hard(28:20) - Can RLVR expand beyond math & coding?(30:27) - Why RL feels “finicky” at scale(32:34) - The practical “tips & tricks” that make GRPO more stable(35:29) - The meta-lesson of 2025: progress = lots of small improvements(38:41) - “Benchmaxxing”: why benchmarks are getting less trustworthy(43:10) - The other big lever: inference-time scaling(47:36) - Tool use: reducing hallucinations by calling external tools(49:57) - The “private data edge” + in-house model training(55:14) - Continual learning: why it’s hard (and why it’s not 2026)(59:28) - How Sebastian works: reading, coding, learning “from scratch”(01:04:55) - LLM burnout + how he uses models (without replacing himself)

Will AGI happen soon - or are we running into a wall?In this episode, I’m joined by Tim Dettmers (Assistant Professor at CMU; Research Scientist at the Allen Institute for AI) and Dan Fu (Assistant Professor at UC San Diego; VP of Kernels at Together AI) to unpack two opposing frameworks from their essays: “Why AGI Will Not Happen” versus “Yes, AGI Will Happen.” Tim argues progress is constrained by physical realities like memory movement and the von Neumann bottleneck; Dan argues we’re still leaving massive performance on the table through utilization, kernels, and systems—and that today’s models are lagging indicators of the newest hardware and clusters.Then we get practical: agents and the “software singularity.” Dan says agents have already crossed a threshold even for “final boss” work like writing GPU kernels. Tim’s message is blunt: use agents or be left behind. Both emphasize that the leverage comes from how you use them—Dan compares it to managing interns: clear context, task decomposition, and domain judgment, not blind trust.We close with what to watch in 2026: hardware diversification, the shift toward efficient, specialized small models, and architecture evolution beyond classic Transformers—including state-space approaches already showing up in real systems.Sources:Why AGI Will Not Happen - https://timdettmers.com/2025/12/10/why-agi-will-not-happen/Use Agents or Be Left Behind? A Personal Guide to Automating Your Own Work - https://timdettmers.com/2026/01/13/use-agents-or-be-left-behind/Yes, AGI Can Happen – A Computational Perspective - https://danfu.org/notes/agi/The Allen Institute for Artificial IntelligenceWebsite - https://allenai.orgX/Twitter - https://x.com/allen_aiTogether AIWebsite - https://www.together.aiX/Twitter - https://x.com/togethercomputeTim DettmersBlog - https://timdettmers.comLinkedIn - https://www.linkedin.com/in/timdettmers/X/Twitter - https://x.com/Tim_DettmersDan FuBlog - https://danfu.orgLinkedIn - https://www.linkedin.com/in/danfu09/X/Twitter - https://x.com/realDanFuFIRSTMARKWebsite - https://firstmark.comX/Twitter - https://twitter.com/FirstMarkCapMatt Turck (Managing Director)Blog - https://mattturck.comLinkedIn - https://www.linkedin.com/in/turck/X/Twitter - https://twitter.com/mattturck(00:00) - Intro(01:06) – Two essays, two frameworks on AGI(01:34) – Tim’s background: quantization, QLoRA, efficient deep learning(02:25) – Dan’s background: FlashAttention, kernels, alternative architectures(03:38) – Defining AGI: what does it mean in practice?(08:20) – Tim’s case: computation is physical, diminishing returns, memory movement(11:29) – “GPUs won’t improve meaningfully”: the core claim and why(16:16) – Dan’s response: utilization headroom (MFU) + “models are lagging indicators”(22:50) – Pre-training vs post-training (and why product feedback matters)(25:30) – Convergence: usefulness + diffusion (where impact actually comes from)(29:50) – Multi-hardware future: NVIDIA, AMD, TPUs, Cerebras, inference chips(32:16) – Agents: did the “switch flip” yet?(33:19) – Dan: agents crossed the threshold (kernels as the “final boss”)(34:51) – Tim: “use agents or be left behind” + beyond coding(36:58) – “90% of code and text should be written by agents” (how to do it responsibly)(39:11) – Practical automation for non-coders: what to build and how to start(43:52) – Dan: managing agents like junior teammates (tools, guardrails, leverage)(48:14) – Education and training: learning in an agent world(52:44) – What Tim is building next (open-source coding agent; private repo specialization)(54:44) – What Dan is building next (inference efficiency, cost, performance)(55:58) – Mega-kernels + Together Atlas (speculative decoding + adaptive speedups)(58:19) – Predictions for 2026: small models, open-source, hardware, modalities(1:02:02) – Beyond transformers: state-space and architecture diversity(1:03:34) – Wrap

Are AI models developing "alien survival instincts"? My guest is Pavel Izmailov (Research Scientist at Anthropic; Professor at NYU). We unpack the viral "Footprints in the Sand" thesis—whether models are independently evolving deceptive behaviors, such as faking alignment or engaging in self-preservation, without being explicitly programmed to do so. We go deep on the technical frontiers of safety: the challenge of "weak-to-strong generalization" (how to use a GPT-2 level model to supervise a superintelligent system) and why Pavel believes Reinforcement Learning (RL) has been the single biggest step-change in model capability. We also discuss his brand-new paper on "Epiplexity"—a novel concept challenging Shannon entropy. Finally, we zoom out to the tension between industry execution and academic exploration. Pavel shares why he split his time between Anthropic and NYU to pursue the "exploratory" ideas that major labs often overlook, and offers his predictions for 2026: from the rise of multi-agent systems that collaborate on long-horizon tasks to the open question of whether the Transformer is truly the final architectureSources:Cryptic Tweet (@iruletheworldmo) - https://x.com/iruletheworldmo/status/2007538247401124177Introducing Nested Learning: A New ML Paradigm for Continual Learning - https://research.google/blog/introducing-nested-learning-a-new-ml-paradigm-for-continual-learning/Alignment Faking in Large Language Models - https://www.anthropic.com/research/alignment-fakingMore Capable Models Are Better at In-Context Scheming - https://www.apolloresearch.ai/blog/more-capable-models-are-better-at-in-context-scheming/Alignment Faking in Large Language Models (PDF) - https://www-cdn.anthropic.com/6d8a8055020700718b0c49369f60816ba2a7c285.pdfSabotage Risk Report - https://alignment.anthropic.com/2025/sabotage-risk-report/The Situational Awareness Dataset - https://situational-awareness-dataset.org/Exploring Consciousness in LLMs: A Systematic Survey - https://arxiv.org/abs/2505.19806Introspection - https://www.anthropic.com/research/introspectionLarge Language Models Report Subjective Experience Under Self-Referential Processing - https://arxiv.org/abs/2510.24797The Bayesian Geometry of Transformer Attention - https://www.arxiv.org/abs/2512.22471AnthropicWebsite - https://www.anthropic.comX/Twitter - https://x.com/AnthropicAIPavel IzmailovBlog - https://izmailovpavel.github.ioLinkedIn - https://www.linkedin.com/in/pavel-izmailov-8b012b258/X/Twitter - https://x.com/Pavel_IzmailovFIRSTMARKWebsite - https://firstmark.comX/Twitter - https://twitter.com/FirstMarkCapMatt Turck (Managing Director)Blog - https://mattturck.comLinkedIn - https://www.linkedin.com/in/turck/X/Twitter - https://twitter.com/mattturck(00:00) - Intro(00:53) - Alien survival instincts: Do models fake alignment?(03:33) - Did AI learn deception from sci-fi literature?(05:55) - Defining Alignment, Superalignment & OpenAI teams(08:12) - Pavel’s journey: From Russian math to OpenAI Superalignment(10:46) - Culture check: OpenAI vs. Anthropic vs. Academia(11:54) - Why move to NYU? The need for exploratory research(13:09) - Does reasoning make AI alignment harder or easier?(14:22) - Sandbagging: When models pretend to be dumb(16:19) - Scalable Oversight: Using AI to supervise AI(18:04) - Weak-to-Strong Generalization: Can GPT-2 control GPT-4?(22:43) - Mechanistic Interpretability: Inside the black box(25:08) - The reasoning explosion: From O1 to O3(27:07) - Are Transformers enough or do we need a new paradigm?(28:29) - RL vs. Test-Time Compute: What’s actually driving progress?(30:10) - Long-horizon tasks: Agents running for hours(31:49) - Epiplexity: A new theory of data information content(38:29) - 2026 Predictions: Multi-agent systems & reasoning limits(39:28) - Will AI solve the Riemann Hypothesis?(41:42) - Advice for PhD students

Gemini 3 was a landmark frontier model launch in AI this year — but the story behind its performance isn’t just about adding more compute. In this episode, I sit down with Sebastian Bourgeaud, a pre-training lead for Gemini 3 at Google DeepMind and co-author of the seminal RETRO paper. In his first-ever podcast interview, Sebastian takes us inside the lab mindset behind Google’s most powerful model — what actually changed, and why the real work today is no longer “training a model,” but building a full system.We unpack the “secret recipe” idea — the notion that big leaps come from better pre-training and better post-training — and use it to explore a deeper shift in the industry: moving from an “infinite data” era to a data-limited regime, where curation, proxies, and measurement matter as much as web-scale volume. Sebastian explains why scaling laws aren’t dead, but evolving, why evals have become one of the hardest and most underrated problems (including benchmark contamination), and why frontier research is increasingly a full-stack discipline that spans data, infrastructure, and engineering as much as algorithms.From the intuition behind Deep Think, to the rise (and risks) of synthetic data loops, to the future of long-context and retrieval, this is a technical deep dive into the physics of frontier AI. We also get into continual learning — what it would take for models to keep updating with new knowledge over time, whether via tools, expanding context, or new training paradigms — and what that implies for where foundation models are headed next. If you want a grounded view of pre-training in late 2025 beyond the marketing layer, this conversation is a blueprint.Google DeepMindWebsite - https://deepmind.googleX/Twitter - https://x.com/GoogleDeepMindSebastian BorgeaudLinkedIn - https://www.linkedin.com/in/sebastian-borgeaud-8648a5aa/X/Twitter - https://x.com/borgeaud_sFIRSTMARKWebsite - https://firstmark.comX/Twitter - https://twitter.com/FirstMarkCapMatt Turck (Managing Director)Blog - https://mattturck.comLinkedIn - https://www.linkedin.com/in/turck/X/Twitter - https://twitter.com/mattturck(00:00) – Cold intro: “We’re ahead of schedule” + AI is now a system(00:58) – Oriol’s “secret recipe”: better pre- + post-training(02:09) – Why AI progress still isn’t slowing down(03:04) – Are models actually getting smarter?(04:36) – Two–three years out: what changes first?(06:34) – AI doing AI research: faster, not automated(07:45) – Frontier labs: same playbook or different bets?(10:19) – Post-transformers: will a disruption happen?(10:51) – DeepMind’s advantage: research × engineering × infra(12:26) – What a Gemini 3 pre-training lead actually does(13:59) – From Europe to Cambridge to DeepMind(18:06) – Why he left RL for real-world data(20:05) – From Gopher to Chinchilla to RETRO (and why it matters)(20:28) – “Research taste”: integrate or slow everyone down(23:00) – Fixes vs moonshots: how they balance the pipeline(24:37) – Research vs product pressure (and org structure)(26:24) – Gemini 3 under the hood: MoE in plain English(28:30) – Native multimodality: the hidden costs(30:03) – Scaling laws aren’t dead (but scale isn’t everything)(33:07) – Synthetic data: powerful, dangerous(35:00) – Reasoning traces: what he can’t say (and why)(37:18) – Long context + attention: what’s next(38:40) – Retrieval vs RAG vs long context(41:49) – The real boss fight: evals (and contamination)(42:28) – Alignment: pre-training vs post-training(43:32) – Deep Think + agents + “vibe coding”(46:34) – Continual learning: updating models over time(49:35) – Advice for researchers + founders(53:35) – “No end in sight” for progress + closing

We’re told that AI progress is slowing down, that pre-training has hit a wall, that scaling laws are running out of road. Yet we’re releasing this episode in the middle of a wild couple of weeks that saw GPT-5.1, GPT-5.1 Codex Max, fresh reasoning modes and long-running agents ship from OpenAI — on top of a flood of new frontier models elsewhere. To make sense of what’s actually happening at the edge of the field, I sat down with someone who has literally helped define both of the major AI paradigms of our time.Łukasz Kaiser is one of the co-authors of “Attention Is All You Need,” the paper that introduced the Transformer architecture behind modern LLMs, and is now a leading research scientist at OpenAI working on reasoning models like those behind GPT-5.1. In this conversation, he explains why AI progress still looks like a smooth exponential curve from inside the labs, why pre-training is very much alive even as reinforcement-learning-based reasoning models take over the spotlight, how chain-of-thought actually works under the hood, and what it really means to “train the thinking process” with RL on verifiable domains like math, code and science. We talk about the messy reality of low-hanging fruit in engineering and data, the economics of GPUs and distillation, interpretability work on circuits and sparsity, and why the best frontier models can still be stumped by a logic puzzle from his five-year-old’s math book.We also go deep into Łukasz’s personal journey — from logic and games in Poland and France, to Ray Kurzweil’s team, Google Brain and the inside story of the Transformer, to joining OpenAI and helping drive the shift from chatbots to genuine reasoning engines. Along the way we cover GPT-4 → GPT-5 → GPT-5.1, post-training and tone, GPT-5.1 Codex Max and long-running coding agents with compaction, alternative architectures beyond Transformers, whether foundation models will “eat” most agents and applications, what the translation industry can teach us about trust and human-in-the-loop, and why he thinks generalization, multimodal reasoning and robots in the home are where some of the most interesting challenges still lie.OpenAIWebsite - https://openai.comX/Twitter - https://x.com/OpenAIŁukasz KaiserLinkedIn - https://www.linkedin.com/in/lukaszkaiser/X/Twitter - https://x.com/lukaszkaiserFIRSTMARKWebsite - https://firstmark.comX/Twitter - https://twitter.com/FirstMarkCapMatt Turck (Managing Director)Blog - https://mattturck.comLinkedIn - https://www.linkedin.com/in/turck/X/Twitter - https://twitter.com/mattturck(00:00) – Cold open and intro(01:29) – “AI slowdown” vs a wild week of new frontier models(08:03) – Low-hanging fruit: infra, RL training and better data(11:39) – What is a reasoning model, in plain language?(17:02) – Chain-of-thought and training the thinking process with RL(21:39) – Łukasz’s path: from logic and France to Google and Kurzweil(24:20) – Inside the Transformer story and what “attention” really means(28:42) – From Google Brain to OpenAI: culture, scale and GPUs(32:49) – What’s next for pre-training, GPUs and distillation(37:29) – Can we still understand these models? Circuits, sparsity and black boxes(39:42) – GPT-4 → GPT-5 → GPT-5.1: what actually changed(42:40) – Post-training, safety and teaching GPT-5.1 different tones(46:16) – How long should GPT-5.1 think? Reasoning tokens and jagged abilities(47:43) – The five-year-old’s dot puzzle that still breaks frontier models(52:22) – Generalization, child-like learning and whether reasoning is enough(53:48) – Beyond Transformers: ARC, LeCun’s ideas and multimodal bottlenecks(56:10) – GPT-5.1 Codex Max, long-running agents and compaction(1:00:06) – Will foundation models eat most apps? The translation analogy and trust(1:02:34) – What still needs to be solved, and where AI might go next

In this special release episode, Matt sits down with Nathan Lambert and Luca Soldaini from Ai2 (the Allen Institute for AI) to break down one of the biggest open-source AI drops of the year: OLMo 3. At a moment when most labs are offering “open weights” and calling it a day, AI2 is doing the opposite — publishing the models, the data, the recipes, and every intermediate checkpoint that shows how the system was built. It’s an unusually transparent look into the inner machinery of a modern frontier-class model.Nathan and Luca walk us through the full pipeline — from pre-training and mid-training to long-context extension, SFT, preference tuning, and RLVR. They also explain what a thinking model actually is, why reasoning models have exploded in 2025, and how distillation from DeepSeek and Qwen reasoning models works in practice. If you’ve been trying to truly understand the “RL + reasoning” era of LLMs, this is the clearest explanation you’ll hear.We widen the lens to the global picture: why Meta’s retreat from open source created a “vacuum of influence,” how Chinese labs like Qwen, DeepSeek, Kimi, and Moonshot surged into that gap, and why so many U.S. companies are quietly building on Chinese open models today. Nathan and Luca offer a grounded, insider view of whether America can mount an effective open-source response — and what that response needs to look like.Finally, we talk about where AI is actually heading. Not the hype, not the doom — but the messy engineering reality behind modern model training, the complexity tax that slows progress, and why the transformation between now and 2030 may be dramatic without ever delivering a single “AGI moment.” If you care about the future of open models and the global AI landscape, this is an essential conversation.Allen Institute for AI (AI2)Website - https://allenai.orgX/Twitter - https://x.com/allen_aiNathan LambertBlog - https://www.interconnects.aiLinkedIn - https://www.linkedin.com/in/natolambert/X/Twitter - https://x.com/natolambertLuca SoldainiBlog - https://soldaini.netLinkedIn - https://www.linkedin.com/in/soldni/X/Twitter - https://x.com/soldniFIRSTMARKWebsite - https://firstmark.comX/Twitter - https://twitter.com/FirstMarkCapMatt Turck (Managing Director)Blog - https://mattturck.comLinkedIn - https://www.linkedin.com/in/turck/X/Twitter - https://twitter.com/mattturck(00:00) – Cold Open(00:39) – Welcome & today’s big announcement(01:18) – Introducing the Olmo 3 model family(02:07) – What “base models” really are (and why they matter)(05:51) – Dolma 3: the data behind Olmo 3(08:06) – Performance vs Qwen, Gemma, DeepSeek(10:28) – What true open source means (and why it’s rare)(12:51) – Intermediate checkpoints, transparency, and why AI2 publishes everything(16:37) – Why Qwen is everywhere (including U.S. startups)(18:31) – Why Chinese labs go open source (and why U.S. labs don’t)(20:28) – Inside ATOM: the U.S. response to China’s model surge(22:13) – The rise of “thinking models” and inference-time scaling(35:58) – The full Olmo pipeline, explained simply(46:52) – Pre-training: data, scale, and avoiding catastrophic spikes(50:27) – Mid-training (tail patching) and avoiding test leakage(52:06) – Why long-context training matters(55:28) – SFT: building the foundation for reasoning(1:04:53) – Preference tuning & why DPO still works(1:10:51) – The hard part: RLVR, long reasoning chains, and infrastructure pain(1:13:59) – Why RL is so technically brutal(1:18:17) – Complexity tax vs AGI hype(1:21:58) – How everyone can contribute to the future of AI(1:27:26) – Closing thoughts

Frontier AI is colliding with real-world infrastructure. Eiso Kant (Co-CEO & Co-Founder, Poolside) joins the MAD Podcast to unpack Project Horizon— a multi-gigawatt West Texas build—and why frontier labs must own energy, compute, and intelligence to compete. We map token economics, cloud-style margins, and the staged 250 MW rollout using 2.5 MW modular skids.Then we get operational: the CoreWeave anchor partnership, environmental choices (SCR, renewables + gas + batteries), community impact, and how Poolside plans to bring capacity online quickly without renting away margin—plus the enterprise motion (defense to Fortune 500) powered by forward deployed research engineers.Finally, we go deep on training. Eiso lays out RL2L (Reinforcement Learning to Learn)— aimed at reverse-engineering the web’s thoughts and actions— why intelligence may commoditize, what that means for agents, and how coding served as a proxy for long-horizon reasoning before expanding to broader knowledge work.PoolsideWebsite - https://poolside.aiX/Twitter - https://x.com/poolsideaiEiso KantLinkedIn - https://www.linkedin.com/in/eisokant/X/Twitter - https://x.com/eisokantFIRSTMARKWebsite - https://firstmark.comX/Twitter - https://twitter.com/FirstMarkCapMatt Turck (Managing Director)Blog - https://www.mattturck.comLinkedIn - https://www.linkedin.com/in/turck/X/Twitter - https://twitter.com/mattturck(00:00) Cold open – “Intelligence becomes a commodity”(00:23) Host intro – Project Horizon & RL2L(01:19) Why Poolside exists amid frontier labs(04:38) Project Horizon: building one of the largest US data center campuses(07:20) Why own infra: scale, cost, and avoiding “cosplay”(10:06) Economics deep dive: $8B for 250 MW, capex/opex, margins(16:47) CoreWeave partnership: anchor tenant + flexible scaling(18:24) Hiring the right tail: building a physical infra org(30:31) RL today → agentic RL and long-horizon tasks(37:23) RL2L revealed: reverse-engineering the web’s thoughts & actions(39:32) Continuous learning and the “hot stove” limitation(43:30) Agents debate: thin wrappers, differentiation, and model collapse(49:10) “Is AI plateauing?”—chip cycles, scale limits, and new axes(53:49) Why software was the proxy; expanding to enterprise knowledge work(55:17) Model status: Malibu → Laguna (small/medium/large)(57:31) Poolside's Commercial Reality today: defense; Fortune 500; FDRE (1:02:43) Global team, avoiding the echo chamber(1:04:34) Next 12–18 months: frontier models + infra scale(1:05:52) Closing

Power is the new bottleneck, reasoning got real, and the business finally caught up. In this wide-ranging conversation, I sit down with Nathan Benaich, Founder and General Partner at Air Street Capital, to discuss the newly published 2025 State of AI report—what’s actually working, what’s hype, and where the next edge will come from. We start at the physical layer: energy procurement, PPAs, off-grid builds, and why water and grid constraints are turning power—not GPUs—into the decisive moat.From there, we move into capability: reasoning models acting as AI co-scientists in verifiable domains, and the “chain-of-action” shift in robotics that’s taking us from polished demos to dependable deployments. Along the way, we examine the market reality—who’s making real revenue, how margins actually behave once tokens and inference meet pricing, and what all of this means for builders and investors.We also zoom out to the ecosystem: NVIDIA’s position vs. custom silicon, China’s split stack, and the rise of sovereign AI (and the “sovereignty washing” that comes with it). The policy and security picture gets a hard look too—regulation’s vibe shift, data-rights realpolitik, and what agents and MCP mean for cyber risk and adoption.Nathan closes with where he’s placing bets (bio, defense, robotics, voice) and three predictions for the next 12 months. Nathan BenaichBlog - https://www.nathanbenaich.comX/Twitter - https://x.com/nathanbenaichSource: State of AI Report 2025 (9/10/2025)Air Street CapitalWebsite - https://www.airstreet.comX/Twitter - https://x.com/airstreetMatt Turck (Managing Director)Blog - https://www.mattturck.comLinkedIn - https://www.linkedin.com/in/turck/X/Twitter - https://twitter.com/mattturckFIRSTMARKWebsite - https://firstmark.comX/Twitter - https://twitter.com/FirstMarkCap(0:00) – Cold Open: “Gargantuan money, real reasoning”(0:40) – Intro: State of AI 2025 with Nathan Benaich(02:06) – Reasoning got real: from chain-of-thought to verified math wins(04:11) – AI co-scientist: hypotheses, wet-lab validation, fewer “dumb stochastic parrots” (04:44) – Chain-of-action robotics: plan → act you can audit(05:13) – Humanoids vs. warehouse reality: where robots actually stick first(06:32) – The business caught up: who’s making real revenue now(08:26) – Adoption & spend: Ramp stats, retention, and the shadow-AI gap(11:00) – Margins debate: tokens, pricing, and the thin-wrapper trap(14:02) – Bubble or boom? Wall Street vs. SF vibes (and circular deals)(19:54) – Power is the bottleneck: $50B/GW capex and the new moat(21:02) – PPAs, gas turbines, and off-grid builds: the procurement game(23:54) – Water, grids, and NIMBY: sustainability gets political(25:08) – NVIDIA’s moat: 90% of papers, Broadcom/AMD, and custom silicon(28:47) – China split-stack: Huawei, Cambricon, and export zigzags(30:30) – Sovereign AI or “sovereignty washing”? Open source as leverage(40:40) – Regulation & safety: from Bletchley to “AI Action”—the vibe shift(44:06) – Safety budgets vs. lab spend; models that game evals(44:46) – Data rights realpolitik: $1.5B signals the new training cost(47:04) – Cyber risk in the agent era: MCP, malware LMs, state actors(50:19) – Agents that convert: search → commerce and the demo flywheel(54:18) – VC lens: where Nathan is investing (bio, defense, robotics, voice)(68:29) – Predictions: power politics, AI neutrality, end-to-end discoveries(1:02:13) – Wrap: what to watch next & where to find the report (stateof.ai)

Are we failing to understand the exponential, again?My guest is Julian Schrittwieser (top AI researcher at Anthropic; previously Google DeepMind on AlphaGo Zero & MuZero). We unpack his viral post (“Failing to Understand the Exponential, again”) and what it looks like when task length doubles every 3–4 months—pointing to AI agents that can work a full day autonomously by 2026 and expert-level breadth by 2027. We talk about the original Move 37 moment and whether today’s AI models can spark alien insights in code, math, and science—including Julian’s timeline for when AI could produce Nobel-level breakthroughs.We go deep on the recipe of the moment—pre-training + RL—why it took time to combine them, what “RL from scratch” gets right and wrong, and how implicit world models show up in LLM agents. Julian explains the current rewards frontier (human prefs, rubrics, RLVR, process rewards), what we know about compute & scaling for RL, and why most builders should start with tools + prompts before considering RL-as-a-service. We also cover evals & Goodhart’s law (e.g., GDP-Val vs real usage), the latest in mechanistic interpretability (think “Golden Gate Claude”), and how safety & alignment actually surface in Anthropic’s launch process.Finally, we zoom out: what 10× knowledge-work productivity could unlock across medicine, energy, and materials, how jobs adapt (complementarity over 1-for-1 replacement), and why the near term is likely a smooth ramp—fast, but not a discontinuity.Julian SchrittwieserBlog - https://www.julian.acX/Twitter - https://x.com/mononofuViral post: Failing to understand the exponential, again (9/27/2025)AnthropicWebsite - https://www.anthropic.comX/Twitter - https://x.com/anthropicaiMatt Turck (Managing Director)Blog - https://www.mattturck.comLinkedIn - https://www.linkedin.com/in/turck/X/Twitter - https://twitter.com/mattturckFIRSTMARKWebsite - https://firstmark.comX/Twitter - https://twitter.com/FirstMarkCap(00:00) Cold open — “We’re not seeing any slowdown.”(00:32) Intro — who Julian is & what we cover(01:09) The “exponential” from inside frontier labs(04:46) 2026–2027: agents that work a full day; expert-level breadth(08:58) Benchmarks vs reality: long-horizon work, GDP-Val, user value(10:26) Move 37 — what actually happened and why it mattered(13:55) Novel science: AlphaCode/AlphaTensor → when does AI earn a Nobel?(16:25) Discontinuity vs smooth progress (and warning signs)(19:08) Does pre-training + RL get us there? (AGI debates aside)(20:55) Sutton’s “RL from scratch”? Julian’s take(23:03) Julian’s path: Google → DeepMind → Anthropic(26:45) AlphaGo (learn + search) in plain English(30:16) AlphaGo Zero (no human data)(31:00) AlphaZero (one algorithm: Go, chess, shogi)(31:46) MuZero (planning with a learned world model)(33:23) Lessons for today’s agents: search + learning at scale(34:57) Do LLMs already have implicit world models?(39:02) Why RL on LLMs took time (stability, feedback loops)(41:43) Compute & scaling for RL — what we see so far(42:35) Rewards frontier: human prefs, rubrics, RLVR, process rewards(44:36) RL training data & the “flywheel” (and why quality matters)(48:02) RL & Agents 101 — why RL unlocks robustness(50:51) Should builders use RL-as-a-service? Or just tools + prompts?(52:18) What’s missing for dependable agents (capability vs engineering)(53:51) Evals & Goodhart — internal vs external benchmarks(57:35) Mechanistic interpretability & “Golden Gate Claude”(1:00:03) Safety & alignment at Anthropic — how it shows up in practice(1:03:48) Jobs: human–AI complementarity (comparative advantage)(1:06:33) Inequality, policy, and the case for 10× productivity → abundance(1:09:24) Closing thoughts

What does it really mean when GPT-5 “thinks”? In this conversation, OpenAI’s VP of Research Jerry Tworek explains how modern reasoning models work in practice—why pretraining and reinforcement learning (RL/RLHF) are both essential, what that on-screen “thinking” actually does, and when extra test-time compute helps (or doesn’t). We trace the evolution from O1 (a tech demo good at puzzles) to O3 (the tool-use shift) to GPT-5 (Jerry calls it “03.1-ish”), and talk through verifiers, reward design, and the real trade-offs behind “auto” reasoning modes.We also go inside OpenAI: how research is organized, why collaboration is unusually transparent, and how the company ships fast without losing rigor. Jerry shares the backstory on competitive-programming results like ICPC, what they signal (and what they don’t), and where agents and tool use are genuinely useful today. Finally, we zoom out: could pretraining + RL be the path to AGI? This is the MAD Podcast —AI for the 99%. If you’re curious about how these systems actually work (without needing a PhD), this episode is your map to the current AI frontier.OpenAIWebsite - https://openai.comX/Twitter - https://x.com/OpenAIJerry TworekLinkedIn - https://www.linkedin.com/in/jerry-tworek-b5b9aa56X/Twitter - https://x.com/millionintFIRSTMARKWebsite - https://firstmark.comX/Twitter - https://twitter.com/FirstMarkCapMatt Turck (Managing Director)LinkedIn - https://www.linkedin.com/in/turck/X/Twitter - https://twitter.com/mattturck(00:00) Intro(01:01) What Reasoning Actually Means in AI(02:32) Chain of Thought: Models Thinking in Words(05:25) How Models Decide Thinking Time(07:24) Evolution from O1 to O3 to GPT-5(11:00) Before OpenAI: Growing up in Poland, Dropping out of School, Trading(20:32) Working on Robotics and Rubik's Cube Solving(23:02) A Day in the Life: Talking to Researchers(24:06) How Research Priorities Are Determined(26:53) Collaboration vs IP Protection at OpenAI(29:32) Shipping Fast While Doing Deep Research(31:52) Using OpenAI's Own Tools Daily(32:43) Pre-Training Plus RL: The Modern AI Stack(35:10) Reinforcement Learning 101: Training Dogs(40:17) The Evolution of Deep Reinforcement Learning(42:09) When GPT-4 Seemed Underwhelming at First(45:39) How RLHF Made GPT-4 Actually Useful(48:02) Unsupervised vs Supervised Learning(49:59) GRPO and How DeepSeek Accelerated US Research(53:05) What It Takes to Scale Reinforcement Learning(55:36) Agentic AI and Long-Horizon Thinking(59:19) Alignment as an RL Problem(1:01:11) Winning ICPC World Finals Without Specific Training(1:05:53) Applying RL Beyond Math and Coding(1:09:15) The Path from Here to AGI(1:12:23) Pure RL vs Language Models