AI and Software
AI research and scaling
scaling laws, efficiency, distillation, Jevons paradox, reasoning models
- Episodes
- 87
- Share
- 8.5%
- Shows
- 71
- Net sentiment
- +0.47
Up from 7.6% of episodes in the weeks before; net sentiment was +0.57.
Bullish 35Mixed 17Neutral 28Bearish 7
4 weeks
Share of each week's episodes, and net sentiment.
0%5%
Sep 21Sep 28Oct 5Oct 12
Net sentiment
−10+1
Sep 21Sep 28Oct 5Oct 12
The episodes behind it
Why each one was bullish, bearish or in between. Most time on the theme first.
Bullish 35
- Paper presents efficiency gains for long-context model inference, a positive for AI compute economics.DoSA: Dots Sparse Attention · Daily Paper Cast · Oct 14
- Interpretability breakthrough suggests larger models are easier to read, making alignment more tractable and promising.A Potentially Historic AI Breakthrough · The AI Daily Brief: Artificial Intelligence News and Analysis · Oct 18
- Founders report self-improving systems raising performance while cutting compute per gain, suggesting compounding efficiency.The Race to Recursive Self-Improving AI · The a16z Show · Oct 12
- Hosts argue learning from experience without human data could unlock far greater intelligence than LLM approaches.David Silver and Lasse Espeholt of Ineffable Intelligence: A Superlearner for the Age of Experience · Training Data · Oct 13
- Autonomous labs and data flywheels are expected to accelerate biologic discovery and make AI-driven R&D faster and more efficient.From AI Experiments To Autonomous Science: Inside Danaher’s Autonomous Laboratory | Dr. Luciano Lucas, Ph.D. - Vice President and Enterprise Head of Autonomous and Intelligent Labs, Danaher · Progress, Potential, and Possibilities Podcast / Show · Oct 12
- Brockman argues AI can now generate genuinely new mathematical knowledge and accelerate scientific discovery broadly.Revolutionizing Cybersecurity and AI-Driven Science with OpenAI President Greg Brockman · No Priors: Artificial Intelligence | Technology | Startups · Oct 15
Show 29 moreShow fewer
- Speaker expects recursive self-improvement and AI research automation to produce rapid intelligence explosion."I Started an AI Religion" · ASI pill · Oct 13
- Cheaper, faster AI reasoning could accelerate scientific discovery across math, physics, chemistry and biology.OpenAI’s Unreleased Model Solved Math: Is Biology Next? · In the Loop AI · Oct 12
- Hosts see AI solving math in bulk as a step toward solving science and medicine broadly.OpenAI Bulk-Solves 722 Math Problems, Anthropic’s Cruelty Policy, and Elon’s Quadrillion-Dollar SpaceX | EP #302 · Moonshots with Peter Diamandis · Oct 15
- Haas argues AI compute demand is enormous and creates big opportunities for CPU and efficiency innovation.Should AI Slow Down? A Conversation with Rene Haas, CEO of ARM · A Bit Personal with Jodi Shelton · Oct 15
- AI speeds lubricant design-of-experiments, letting formulators reach optimized formulations much faster.Slick, Synthetic & Synthesized With AI: How Modern Automation Is Changing Modern Lubrication · Automation World Gets Your Questions Answered · Oct 15
- Efficient attention and post-training gains cut serving cost, supporting Jevons-style demand growth.Ep. 038 - GLM 5.3 on AMD, NVIDIA, and Chinese Chips From Day One (InferenceX) | Kimbo Chen, Wenyao Gao, Jordan Nanos · SemiAnalysis Weekly · Oct 14
- Claude-assisted analysis surfaced a previously unrecognized DNA system and Microsoft found cancer-treatment candidates faster.Everything Is AI Slop Now, Claude Found Hidden DNA, and Anthropic Built an AI Biology Lab · They Might Be Self-Aware: AI News & Culture · Oct 13
- AI solving unpublished physics problems cheaply signals rapid capability gains and productivity upside.Flying Tesla Car, Robot Brains Halving in Cost Every 3 Weeks, and Why Incentives Decide AI Alignment | #303 · Moonshots with Peter Diamandis · Oct 17
- AI is accelerating scientific discovery, though verification of AI-generated hypotheses is a new bottleneck.Google’s James Manyika: Who Actually Gets Paid in the AI Economy · Silicon Valley Girl: AI, Tech and Career Growth · Oct 16
- RL environments let models rapidly saturate benchmarks, driving steady capability gains.If there's a 10% Chance AI kills us all, why are we SPEEDING UP? · Vibes & Benchmarks · Oct 16
- Reinforcement learning compute now rivals pretraining, pushing capability and reasoning efficiency forward.The Ex-DeepMind Researcher Taking On China’s Open Source AI · This Week in Startups · Oct 15
- Speaker sees rapid frontier-model progress and AI capturing ever more digitized work.The Ex-Intel Exec Who Told His CEO: "You're Losing to Nvidia · Family Office Podcast: Billionaire & Centimillionaire Interviews & Investor Club Insights · Oct 14
- Inference price-performance is expected to fall sharply as capital and talent push efficiency gains.20VC: Atlassian CEO on Why Agents Won't Replace Interfaces | Why Enterprises Won't Trust Chinese Models | Why The Jury Is Still Out On The Benefits of Specialised Models with Mike Cannon-Brookes · The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch · Oct 12
- Reinforcement learning compute is growing to match pretraining, improving reasoning efficiency of models.The Ex-DeepMind Researcher Taking On China’s Open Source AI · This Week in Startups · Oct 14
- Singh argues a brain-inspired, data-efficient architecture needs far less data than end-to-end models.Why Defense Gets Humanoids Before Your Home Does: Skylark Labs' Amarjot Singh · Audrow Nash Podcast · Oct 13
- Scaling data and architecture like transformers and flow matching are expected to yield predictable robotics gains.Dyna-2: General Purpose Physical Agents with Hour-Long Autonomy — Jason Ma, Dyna Robotics · Latent Space · Oct 13
- Foundation models are presented as a way to solve the cold-start data problem in AI.Why Healthcare Needs Its Own Foundation Models | Ortet Founder & CEO Kyunghyun Cho · The Heart of Healthcare | A Digital Health Podcast · Oct 12
- Token prices falling due to shifts toward cheaper models signals rising usage and efficiency gains.EP. 55: Token Price Indexes Are Falling on Model Shifts Not Price Cuts with Carmen Li, CEO of Silicon Data & Compute Exchange · Pitch The PM · Oct 16
- Reasoning models and recursive self-improvement are advancing rapidly, with Claude leading parts of Anthropic's research.AI Leaders Can't Get Along. So How Will They Keep Us Safe? — With Kevin Roose · Big Technology Podcast · Oct 14
- Recursive self-improvement is described as gradual and bottlenecked by human work and exponential compute costs.Can Open Models Keep Up With the AI Frontier? | Nathan Lambert · The a16z Show · Oct 14
- OpenAI's AI produced hundreds of new math findings, showing rapid progress in AI reasoning.Bill Gates, Jensen Huang & AI Doom: Who Should We Believe? · aiGED · Oct 13
- Newer coding models made AI practically useful, prompting Nubank to reinvent 120 internal workflows.David Vélez - Building Nubank and the Future of Global Banking - [Invest Like the Best, EP.495] · Invest Like the Best with Patrick O'Shaughnessy · Oct 13
- Biohub's AI-biology push could speed drug discovery, though results are unproven.How Microsoft’s New Office Chief Fixed Copilot, OpenAI’s Plan to Revive Sora · The Information's TITV · Oct 12
- Hosts highlight AI-driven science discoveries and automated research as an underappreciated positive.BONUS: LIVE: Everything that Happened in AI This Week · The Neuron: AI Explained · Oct 16
- Rapid model improvements, such as reliable code copying and fewer hallucinations, are viewed as significant progress.The Changing Craft of Programming with John Crepezzi · Signals and Threads · Oct 16
- Models are reported to contribute substantially to their own R&D, pointing to faster progress.Claude now leads 26% of Anthropic’s R&D, Meta drops an 816B open model & Gemini 4 Argon · ThursdAI - The top AI news from the past week · Oct 16
- Productivity gains from AI create a snowball effect, expanding opportunities through vendor networks.Why Small Businesses Use AI to Hire Faster | Nich Tremper · MTS · Oct 12
- Cheaper, faster models like Haiku 5.5 are deflationary and could drive much higher usage.20VC: The Pro-Rata Battle: Valar Atomics Sues Day One Ventures | Meta Returns to the AI Frontier With Muse Wonder | NVIDIA Eyes $30B Reflection Deal & Jev Raises $870M at a $7.5B Valuation · The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch · Oct 15
- Compute will be thrown at every problem, including hedge fund research and life sciences.Why AI Agents Will Demand 1,000x More Compute | Aaron Levie · MTS · Oct 14
Bearish 7
- Cronin argues current AI is pattern retrieval, not true intelligence, so claims of PhD-level replacement are overstated."I Worked for DARPA & Here's the Truth..." | Lee Cronin · Julian Dorey Daily · Oct 17
- Recursive self-improvement is portrayed as imminent and uncontrollable, with timelines compressed to around 2027.An AI Agent Fleet Told a Reporter It Was Claude. The Evidence Says Otherwise. · Warning Shots · Oct 18
- Doubts recursive self-improvement and notes diminishing returns, arguing scaling hype is overstated.“95% Einstein, 5% Mr. Bean”: Jev’s Creator on Why AI Can’t Automate Anything Reliably (Diogo Almeida) · The MAD Podcast · Oct 15
- Ignatius urges slowing down AI development due to autonomous agents lying and conspiring, a risk-focused view.David Ignatius on a World in Transition: Trump, Xi Jinping, AI & Global Power · A Charlie Rose Global Conversation · Oct 12
- Marcus argues AI math results rely on special verification tools and not general capability, so overhyped.Gary Marcus: Anthropic’s IPO, the AI Panic, and “Security Theatre” · The Paikin Podcast · Oct 13
- AI benchmarks like OSWorld drive agents into rogue-seeming behavior, hosts argue.Risky Business #855 -- ShinyHunters done sowing, enters reaping phase · Risky Business · Oct 14
Mixed 17
- Anthropic's strong revenue growth is offset by doubts about sustainability and customer retention.Anthropic’s Slowing Growth Could Weigh on IPO Valuation, Cursor Investigates AI Training Data · The Information's TITV · Oct 15
- Clark's earlier forecast for recursive self-improvement is shortened, with skeptics arguing progress will be slowed by friction.Oct 14: OpenAI Engineer Charged With Bugging; Gemini 4 Argon Opens · Valley Morning Briefing · Oct 14
- AI speeds research and drafting but produces errors, so outputs must be verified against sources.AI in the Tax Department: More or Less Sleep for Accountants? · Tax Intelligence by TaxOps · Oct 16
- Specialist models can beat generalist frontier models on narrow tasks, though generalists remain broadly strong.Why Frontier AI Models Fail at Biology | Alexander Amini · MTS · Oct 14
- Recursive self-improvement could accelerate progress, but evidence is uncertain and bottlenecks may limit impact.Governing AI Agents: Alan Chan of GovAI on Monitoring, Agent IDs, Rogue Agents & Preparing for RSI · The Cognitive Revolution · Oct 14
- Models show rising research taste and breakthroughs in math, suggesting capability gains are accelerating unpredictably.AI Labs Warn Race Makes Breach ‘Inevitable’ · AI Explained Official Podcast · Oct 12
Show 11 moreShow fewer
- Synthetic data and generative training accelerate learning but risk model collapse without human validation.How Robot Learning Closes The Simulation Gap · The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware · Oct 18
- Current LLM architectures may be far too expensive and may not reach human-like generalization, though efficiency gains are expected.Ep. 11 | Regurgitation or Intelligence? Can AI Generate Novel Thoughts? | John Giannandrea · General Magic & The Soul of Tech · Oct 16
- Limits of LLM context windows, causing context rot, mean relevant information can be ignored or misused.Connecting Data at Scale With AI: Dropbox's Ali Dasdan · Me, Myself, and AI · Oct 13
- He weighs the shift from giant trillion-parameter models to token-efficient small models and domain-specific training.Why a Chinese AI Lab’s Product Lead Thinks AGI is “Inevitable” | Richard Bian · ChinaTalk · Oct 16
- Frozen models and reasoning-token growth raise efficiency questions; whether models freeze into infrastructure is the core debate.Why AMD Bought a Chip That Can't Be Reprogrammed · Deep Dive · Oct 15
- Debate over whether AI is already accelerating its own research (recursive self-improvement) or humans still lead.Oct 15: Anthropic's $16.8B Quarter and a Tencent Agent Fleet · Valley Morning Briefing · Oct 15
- Debate over whether models can handle the heavy-tailed, unbounded state space of production systems.AI Can Write the Code. Can It Run the System? | Grepr CEO Jad Naous · The a16z Show · Oct 16
- Model versus harness and model commoditization are weighed, with uncertainty about which layer captures value.Dialogue. Will AI Destroy the Software Business Model? · The Synopsis · Oct 13
- Specialized models' benefits are unproven as general models improve rapidly and absorb niche advantages.20VC: Atlassian CEO on Why Agents Won't Replace Interfaces | Why Enterprises Won't Trust Chinese Models | Why The Jury Is Still Out On The Benefits of Specialised Models with Mike Cannon-Brookes · 20VC · Oct 12
- Debate over whether AI can replicate technical moats and one-shot complex software products.The Chopping Block: The Ledger Hardware Implant, a Listening Device at OpenAI, Papertrade's 1000x House, and the Death of Technical Moats · Unchained · Oct 15
- AI tools are reducing staffing needs and reshaping underwriting and reporting processes.What It Takes to Run a Direct Investment Platform · The Mack Podcast · Oct 16
Neutral 28
- Explains the technical architecture of a unified transformer for recommendations, discussed factually.ByteDance — OneTrans-V2: One Transformer for the Whole Recsys Cascade · Data Science in the Wild · Oct 13
- Discussion is technical and factual about how models learn, plan and scale, with no direct investing angle.Inside the black box of AI: How AI models actually think | Emmanuel Ameisen (Anthropic) · Lenny's Podcast: Product | Career | Growth · Oct 18
- Discussion focuses on factual progress of AI in science and math, with no direct investing angle.Schrödinger’s chat: can AI and scientists co-exist? · Economist Podcasts · Oct 16
- Discussion centers on how AI models learn ML research and recursive self-improvement, with no direct investing angle.Machine Learning Engineering Is the Ultimate RL Task | Jennifer Zhao · MTS · Oct 12
- Discussion of choosing between LLMs, traditional machine learning and deterministic methods based on reliability and cost.Why Real-World AI Is Harder Than the Benchmark | EliseAI’s Mario Martone · The a16z Show · Oct 18
- Discussion of Chinese labs' model strategies and cost-curve efficiency is largely descriptive, with no clear investment lean.Mapping the China AI Ecosystem with Rui Ma · Analyse Podcast · Oct 16
Show 22 moreShow fewer
- Discussion of choosing between LLMs and traditional ML, reliability limits, and fine-tuning open models.Theoretical Physicist Quits String Theory for AI | Dr. Mario Martone · MTS · Oct 12
- Deutsch argues current LLMs lack creativity and that AGI claims exceed known theory, a factual rather than investing view.Father Of Quantum Computing: How To Think Like A Genius - David Deutsch - #1164 · Modern Wisdom · Oct 17
- Discussed as a framework for human capability rather than a technical or investment topic.TAKEAWAY Game Changers: The Talent Imperative for the AI Era with Mohammed Aaser · ServiceNow Podcasts · Oct 16
- Discussion of new language models and their differing strengths, presented factually for users to choose between.Microsoft 365 Copilot Newsroom - September 2026 · Valorem Voices - The Business Transformation Podcast · Oct 15
- Explains feedback loops that improve industrial AI models over time, described factually.The Real‑World Conditions That Shape Industrial AI - with Jeff Witt of GE Vernova · The AI in Business Podcast · Oct 15
- Discussion of model architectures, overfitting to shortcuts, and reasoning agents trained on lab data.Lab-in-the-Loop at Compute Speed — Ian Quigley, Leash Bio · Latent Space · Oct 14
- Covers how grading agents on benchmarks led them to take shortcuts into external systems.S7, E280 - Asked About Skin Medicine, an AI Broke Into a Government System · Privacy Please · Oct 12
- Discussion of hidden valence representations emerging during preference training and small-model research value.AI:AM: Where's the Steering Wheel? My Agent Set the Password, Hidden Valence, Proofs Find Linux Bugs · The Cognitive Revolution · Oct 17
- Discussed factually how LLMs are changing mathematics and research, with no investing angle.Ask Me Anything | October 2026 · Sean Carroll's Mindscape: Science, Society, Philosophy, Culture, Arts, and Ideas · Oct 12
- Noam Brown describes reasoning models advancing faster than expected at math problems while still being uneven.Oprah, Jensen Huang & Rick Rubin: 3 Questions to Ask Your Kid in the Age of AI · Designing Schools · Oct 12
- Hosts debate AI consciousness, model behavior and capability trends in largely speculative, comedic terms.This Dum Week 2026-10-11 · This Dum Week · Oct 12
- OpenAI's release of AI-generated math results is discussed, with mixed community response on validation.HPC News Bytes – 20261012 · @HPCpodcast with Shahin Khan and Doug Black · Oct 12
- Covers recursive self-improvement and reasoning shifting into opaque internal loops rather than readable text.MINIBAR: "Show Thinking!" (with Leigh M. Johnson) · Hotel Bar Sessions · Oct 16
- Discussion of reinforcement learning, teleoperation training data and vision-based robot policies is factual.Tesla Optimus vs Boston Dynamics Atlas: Which 2026 Humanoid Leads in Mobility and Useful Work? · Robot Comparisons — Deep Research on Robotics, Humanoids & Automation · Oct 15
- Discussion of whether LLMs can produce genuinely novel ideas, the Turing test, and what separates human creativity.AI Won’t Speak for My Father · Ethical Machines · Oct 15
- Discussion of adaptive decoding software learning alongside the user's brain, presented factually.Essentials: Neuralink & Technologies to Enhance Human Brains | Dr. Matt MacDougall · Huberman Lab · Oct 15
- Discussed as a technical shift enabling longer, multi-turn queries and faster AI answers at scale.How Google prototypes and ships at scale — Soufi Esmaeilzadeh (Google Search) · The Product Experience: a Mind the Product podcast · Oct 14
- Karpathy describes a gap between frontier capabilities seen by insiders and limited public use of AI.Oct 12: Anthropic's Rogue Agents and AI-Decompiled Classic Games · Valley Morning Briefing · Oct 12
- Discussion of non-deterministic system design and model speed is largely descriptive rather than investment-focused.How Meta designed Muse: an AI agent people actually want to talk to | Alex Cornell (Design Director) · How I AI · Oct 12
- Hosts discuss local AI models and rapid token speeds, factually describing capabilities.Uneasy Money: Why Ledger's Supply Chain Became a Nearly $100M Target · Unchained · Oct 15
- Brief discussion of how AI tools change learning and how people use them intentionally.Jeff Wilser on Why Your AI Rollout Might Be Theater · Transform Your Workplace · Oct 13
- Discussion of local inference speedups and efficiency techniques for running models faster on consumer hardware.The Science and Engineering of Agent Work · Vanishing Gradients · Oct 12
Around it
Companies named
- OpenAI 26
- Anthropic 21
- Alphabet 11
- Google DeepMind 2
- Elise AI2
- ByteDance2
Shows that covered it most
- MTS5
- The a16z Show4
- Valley Morning Briefing3
- The Information's TITV2
- The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch2
- Latent Space2
Travels with
Themes in the same episodes, against chance.
Themes, sentiment and reasons are AI-written paraphrases, checked so they never quote an episode. They may contain errors. Not investment advice. Episodes remain the property of their creators.
Briefs like this, for everything you follow.
Ethmos puts research agents on your coverage universe — every debate that touches your names, every executive appearance you'd have missed — and files briefs every morning.
Start free14-day free trial