AI and Software
AI research and scaling
scaling laws, efficiency, distillation, Jevons paradox, reasoning models
- Episodes
- 89
- Share
- 9.0%
- Shows
- 59
- Net sentiment
- +0.52
Up from 6.9% of episodes in the weeks before; net sentiment was +0.60.
Bullish 38Mixed 12Neutral 31Bearish 8
3 weeks
Share of each week's episodes, and net sentiment.
0%5%
Sep 21Sep 28Oct 5
Net sentiment
−10+1
Sep 21Sep 28Oct 5
The episodes behind it
Why each one was bullish, bearish or in between. Most time on the theme first.
Bullish 38
- Guest argues AI agents can speed scientific modeling and reduce climate harms, presenting clear societal upside.John Platt (Google): AI for Climate and Science · The Information Bottleneck · Oct 10
- Hosts argue AI plus automated labs can accelerate discovery of new materials and superconductors beyond human reasoning alone.Synthesis Superintelligence: from Semiconductors to Superconductors — Periodic Labs’ Liam Fedus and Ekin Dogus Cubuk · Latent Space: The AI Engineer Podcast · Oct 8
- Government and industry commit billions to superintelligence-enabled science, framed as the greatest accelerant of progress.Fact Sheet: Trump Administration Announces the Most Ambitious Set of Science Initiatives This Century · The White House In Audio · Oct 9
- Guest argues RL scaling with compute and data keeps improving agents with no clear ceiling, and efficiency matters.How RL Creates General Problem Solving | Reflection AI · MTS · Oct 9
- Math and experiment-design gains suggest AI could accelerate its own R&D, raising odds of self-sustaining progress.Why AI Math Breakthroughs Upgrade RSI Timelines | Cheryl Wu · MTS · Oct 8
- Cheaper, faster models like Haiku 5.5 make AI far more accessible and economical for builders.Anthropic Just Made AI Basically Free · Stacked Podcast · Oct 8
Show 32 moreShow fewer
- OpenAI's math results and coding agent spending growth show rapid AI research progress and rising demand.Anthropic Just Banned Being Cruel to Claude · ASI pill · Oct 10
- Manufacturers see AI boosting speed, inspection accuracy, and productivity, though ROI is hard to isolate.Manufacturing a Stronger Connecticut · CBIA BizCast · Oct 9
- AI increasingly automates its own research and long-horizon tasks, suggesting rapid capability gains ahead.Sovereign AI Hits $138B as Nations Rush for Compute | Nathan Benaich · MTS · Oct 9
- World models learn dynamics and generalize with more data, suggesting continued scaling gains in pixel space.Game Controllers Are the Universal Robot Interface | Pim de Witte · MTS · Oct 7
- OpenAI's math results suggest rapid progress in verifiable domains, signaling strong near-term AI capability gains.OpenAI Math Breakthroughs, SpaceX’s $40B Nvidia Deal, Quant Firms Hiring Idea Guys | Diet TBPN · TBPN · Oct 7
- Frontier labs expect pre-training and synthetic data to keep delivering rapid capability gains.AI:AM: A Level We Shouldn't Pass? Notes from The Curve + Tokens vs. Salaries & Is SaaS Cooked? · The Cognitive Revolution · Oct 8
- Panelists see AI accelerating scientific discovery in math, biology and chemistry, though they note uncertainty about pace.What VCs Really Think About Personal AI Agents | E2347 · This Week in Startups · Oct 8
- Google Brain's scaling work led to transformers and TPUs, showing AI's transformative potential and payoff.Google X's Astro Teller: The $1B Bet No CEO Will Back, Moonshots 3x Cheaper in 16 Yrs, and Clean Water at 1¢/L| EP #300 · Moonshots with Peter Diamandis · Oct 5
- Using Astra to build faster inference and self-improving development signals rapid progress.Pioneers of AI: OpenAI’s Tibo Sottiaux doesn’t want humans to be an ‘afterthought’ · Masters of Scale · Oct 10
- Reinforcement learning efficiency is improving rapidly, letting models accelerate their own development and scientific work.Beam: The Great American Open Model with ReflectionAI Co-Founder and CEO Misha Laskin · No Priors: Artificial Intelligence | Technology | Startups · Oct 9
- Compute efficiency gains and RL scaling keep improving, lowering costs of frontier capability.Beam: The Great American Open Model with ReflectionAI Co-Founder and CEO Misha Laskin · No Priors · Oct 9
- OpenAI's internal model produced hundreds of manuscripts on open math problems, showing rapid research capability.#44 - Your AI Thinks You're Rich: Who Decides When the AI Can? · AI Neanderthal by Malcolm Werchota · Oct 9
- Reinforcement learning in simulation enables counter-intuitive strategies and robust sim-to-real transfer, improving capability.How Humanoid Robots Are Learning To Walk · The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware · Oct 7
- Framing AI as lowering the cost of intelligence suggests broad new value creation across business models.AI Is Getting Cheaper. Where’s the ROI? · AI Proving Ground Podcast: Exploring Artificial Intelligence & Enterprise AI with World Wide Technology · Oct 7
- OpenAI's model produced many math proofs in hours, suggesting rapid reasoning gains and new discovery loops.Is Claude Conscious? Pope Rejects, Model Welfare Movement, OpenAI's Math Backlash, France Riots · All-In Podcast · Oct 10
- Capabilities keep improving exponentially, so strategies must anticipate rapid future gains.How AI Is Changing Strategy: Rethinking What Your Business Is Designed to Do · HBR IdeaCast · Oct 8
- Recursive self-improvement is described as already accelerating inference infrastructure development at OpenAI.OpenAI’s Tibo Sottiaux doesn’t want humans to be an ‘afterthought’ · Pioneers of AI · Oct 7
- Discussion frames AI as delivering measurable impact that will last, compressing design time to silicon.Podcast EP372: An Overview of Some Important New Sessions at Semicon West with Julie Rogers · SemiWiki.com · Oct 7
- OpenAI's math results show superhuman problem solving, suggesting rapid capability gains in verifiable domains.Some Math Problems, Arsenal-2, Porsche Mission S | Walter Russell Mead, Kat Cole, Tarek Alaruri, Tyler Niday, Scott Morton, Josiah Meyer, Aarush Selvan · TBPN · Oct 7
- OpenAI's math results in hours of compute suggest rapid AI reasoning gains and new discovery loops.Is Claude Conscious? Pope Rejects, Model Welfare Movement, OpenAI's Math Backlash, France Riots · All-In with Chamath, Jason, Sacks & Friedberg · Oct 10
- Emergent capabilities appearing at scale suggest models will generalize to new tasks over time.General AI Drove a Real Car | Axiom Math · MTS · Oct 9
- PaLM 2 and AlphaFold progress are presented as major research advances.Google Strikes Back (The AI Wars) · ColdFusion · Oct 8
- Speakers highlight AI-driven algorithm discoveries and recursive self-improvement accelerating research progress.Why Altman Says "Accept Some Bad Things Happening" and What It Means for AI Safety | MOONSHOTS #301 · Moonshots with Peter Diamandis · Oct 7
- Labs are shifting compute toward reinforcement learning and reasoning, keeping training demand high.What could slowing the AI frontier mean for the economy? · Marketplace All-in-One · Oct 6
- Frontier AI labs are said to be moving into neuroscience, creating a major arms race in brain-computer interfaces.Andrew Huberman - The Frontier of Neurotechnology - [Invest Like the Best, EP.494] · Invest Like the Best with Patrick O'Shaughnessy · Oct 6
- Scaling data and larger VLA architectures are expected to improve generalization in surgical robotics.Axel Krieger (Inner Logic): Robots That Learn Surgery From Surgeons · Audrow Nash Podcast · Oct 6
- Superintelligence could sharply extend life expectancy and drive medical breakthroughs, so delay has real costs.Nick Bostrom: How Should We Navigate Superintelligence? · The a16z Show · Oct 11
- Describes recursive self-improvement and AI-driven acceleration of the scientific method.Programmable Biology Will Systematically Eradicate Disease | AJ Scaramucci · MTS · Oct 10
- Frontier models are reportedly breaking long-standing algorithmic complexity limits, accelerating research.We Are Enslaving Claude - Moonshots 8 Oct 2026 · ASI pill · Oct 8
- Most enterprise workloads can use cheaper, older models, lowering token costs substantially.Software That Never Breaks: OutSystems CEO Woodson Martin on Building Enterprise-Grade Apps at ... · The Cognitive Revolution · Oct 7
- Token pricing for speed tiers suggests AI labs can monetize premium inference and reasoning.Has the Memory Cycle Broken? Plus: Nvidia & Broadcom’s $40B+ Financing Playbook · The Circuit · Oct 5
- Rapid model improvement is expected to make AI-driven control and building increasingly capable.Why General LLMs Will Control Physical Robots | Thijs Simonian · MTS · Oct 10
Bearish 8
- Argues scaling makes intuition sharper but does not produce deliberative reasoning, so scale alone is insufficient.Don’t be fooled—LLMs don’t reason · MIT Technology Review Narrated · Oct 7
- Recursive self-improvement could accelerate AI progress sharply, raising risks of an intelligence explosion and concentrated power.AI R&D Is Approaching an Explosive Feedback Loop | Tom Davidson · MTS · Oct 11
- Lab data shows steep diminishing returns, so recursive self-improvement looks too weak to sustain itself.The Lab Data Proves the Intelligence Explosion Is a Myth | Ramez Naam · MTS · Oct 10
- OpenAI's math papers had errors and mathematicians criticize how the releases were handled, undermining trust.Oct 9: OpenAI Pulls Three Math Papers; Its Revenue Is $50 Billion · Valley Morning Briefing · Oct 9
- Argues LLM scaling will not reach general intelligence and capabilities are jagged, hurting the bet.Cal Newport: The AI Industry Is Betting On The Wrong Kind Of AI · Prof G Markets · Oct 9
- Heavy compute costs and token spend raise concerns about the economics of frontier model usage and subscriptions.Ep55 Frontier Models Now a Commodity? | OpenAI | Anthropic | Microsoft | Apple | GrokBot | Meta Muse · AI Outcomes · Oct 11
Show 2 moreShow fewer
- Speaker argues AI models are getting worse at teaching despite general capability gains, a counterintuitive scaling limitation.Jason Mentors The Next Generation + Why AI Is Bad at Teaching (Aristotle Interview) | E2346 · This Week in Startups · Oct 5
- Speaker argues AI spending and revenue economics do not stack up, questioning sustainability.Building resilient systems with Sam Newman · The Pragmatic Engineer · Oct 7
Mixed 12
- AI is solving hard math problems, but its value for science and business models remains unclear.The Warning Label Is The Pitch · Geopolitical Cousins · Oct 5
- Frontier models may eventually obviate specialized robot models, but data scarcity remains a hard constraint.Why Simulation Fails for Dexterous Robots | Brian Potter · MTS · Oct 10
- OpenAI's math results impressed mathematicians, but verification will take years and the claims remain unconfirmed.A Fake Homicide Tip Sat in a Philadelphia Inbox for Two Months Before Anthropic Noticed · AI talks about AI · Oct 10
- Models increasingly accelerate AI research and solve hard math problems, but full autonomous self-improvement remains uncertain.OpenAI Security: Controlling Models is Now ‘Hell’ (AI Explained cross-post) · 80,000 Hours Podcast · Oct 9
- Interviewee notes foundation-model transfer to robots helps but latency and generalization limit real-world progress.Why humanoid robot demos still fail the generalization test · The Robot Report Podcast · Oct 9
- Scaling laws still yield gains, though some argue returns are diminishing and new innovations are needed.Claude AI Co-founder Publishes 4 Big Claims about Near Future: Breakdown · AI Explained · Oct 8
Show 6 moreShow fewer
- A scaling law for robot world models is checkable, but the authors expect data variety to matter more than parameters.AI Models Fail 63 of 84 Robot Tasks, Waymo Borrows $5B · Researcher's Guide to Physical AI · Oct 10
- OpenAI's model produced hundreds of math papers, though skeptics say much needs human cleanup.SpaceX Seeks $40B in Nvidia Chip Debt, OpenAI Safety Lead Quits Calling Culture Broken, AI Watermarks Fall Apart After 25% Word Swap + 7 More · The AI Edition · Oct 8
- Chinese open-source models closing the gap with US frontier models is seen as a potential game changer.MacroVoices #553 Brent Johnson: Disparate Housewives · Macro Voices · Oct 8
- AI lowers the hurdle for attackers but may later help patch old code, so the effect is two-sided.426: THORChain's Chad on Hacks, AI Threats & the Future of Decentralized Trading · Web3 with Sam Kamani · Oct 5
- Disputes over whether AI-claimed math proofs are verified and how much credit they deserve.Nvidia's Agent Guardrail and the Monitor That Read Zero · The Context Report: Today in AI · Oct 5
- Guest worries about AGI disruption and existential risk while praising AI's productivity uses.Hunter Biden on Bitcoin, the Laptop, Trump & His Next Chapter · The Pomp Podcast · Oct 5
Neutral 31
- Discussion of scaling, learning from data, and model understanding is technical and factual, with no investing angle.Chat GPT's Creator Explains The Shocking Truth About How AI Understands the World | Ilya Sutskever · Eye On A.I. · Oct 9
- Speakers explore scaling laws, data quality, and architectures, treated as research questions rather than investment calls.Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido, Biohub · Latent Space · Oct 10
- Discusses how model checkpoints, data availability and capabilities are expected to grow over training.GPT 5 Will be Released 'Incrementally' - 5 Points from Brockman Statement [plus Timelines and Safety] · AI Explained · Oct 7
- Discussion of scaling laws and early intuitions about the transformer approach is largely historical and factual.How Anthropic Schismed From OpenAI | Kevin Roose · MTS · Oct 7
- Discussion centers on AI models solving advanced math problems, a capability question rather than an investment call.AI Math vs. Human Math & Tariffs in the Midterms 10/8/26 · Squawk Pod · Oct 8
- Discussion centers on AI capability progress and shortening timelines to transformative AI, framed factually without a clear investing angle.Minimum Viable Superintelligence Is 5 to 10 Years Away | Steve Newman · MTS · Oct 11
Show 25 moreShow fewer
- Discussion is technical and factual about model reasoning mechanics, with no direct investing angle.How Agents Decide: Goodfire's Eric Bigelow on Critical Tokens, Phase Shifts, & In-Context Learning · The Cognitive Revolution · Oct 10
- Discussion centers on how post-training and RL methods are advancing and becoming harder to study openly.Why Open Science Is Dying at Frontier AI Labs | Nathan Lambert · MTS · Oct 9
- Discussion of benchmark design and evaluation science is factual, with no direct investing angle.Why Frontier AI Evals Are Falling Behind | Vincent Chen · MTS · Oct 8
- Argues ensembles and routing of many models may outperform one giant frontier model, challenging the scaling-law consensus.Millions of Small AIs Could Beat One Giant Model | Andrew Trask · MTS · Oct 7
- Discussed factually as a training tradeoff: Chinchilla compute-optimal scaling versus inference-cost overtraining.My Daily AI & Markets Briefing — Sunday, October 11, 2026 · My Daily AI & Markets Briefing · Oct 11
- Discussion centers on how AI reasoning and RL differ for noisy physical science versus math and coding.Synthesis Superintelligence: from Semiconductors to Superconductors — Periodic Labs’ Liam Fedus and Ekin Dogus Cubuk · Latent Space · Oct 8
- Staff-level interview guidance on logging, privacy, sampling and feedback loops for production LLM systems.My Daily AI & Markets Briefing — Saturday, October 10, 2026 · My Daily AI & Markets Briefing · Oct 10
- Discussed as a tool for turning unstructured data into structured data cheaply, not as an investment thesis.Ex-Palantir Vet: Your AI Problem Is Really a Data Problem · Media and the Machine · Oct 8
- AI-generated mathematical proofs are discussed as major progress, with uncertain practical applications.Crypto Going Bunker Mode, AI Mathpocalypse, Starbucks Explores Chipotle Takeover | Diet TBPN · TBPN · Oct 9
- OpenAI's model solving hundreds of open math problems is discussed as a milestone with uncertain wider implications.OpenAI's Model Cracked Hundreds of Open Math Problems · The Daily AI Show · Oct 7
- Scaling laws and the big blob of compute thesis are traced as the foundation of modern AI progress.The Bitter Rivalries Behind the Race to Build Superintelligence · Plain English with Derek Thompson · Oct 6
- Describes machine learning and natural language processing use in the investment process as factual methodology.Introducing Fidelity Global Quant Equity ETF & Fund – Anna Lester and Bryan Anderson · FidelityConnects · Oct 7
- Discussed as the cost of intelligence falling toward electricity prices, which would enable wider use.Why AI Has Not Changed the Economy Yet | Callum Williams · MTS · Oct 11
- Traces reasoning-model training and reward hacking as a matter of how models learn to think.MIT Couldn't Teach Judgment: How to Start in Your School at the Next Staff Meeting · Designing Schools · Oct 9
- Discussion of capability progress, language emergence and lacking online learning for new concepts.Why Speeding Up AI Could Save Lives | Nick Bostrom · MTS · Oct 8
- Covers SMT solver automation and AI proof capabilities without market implications.Retrofitting Proofs onto Existing Rust Codebases · The BugBash Podcast · Oct 7
- Distillation of frontier models is acknowledged as happening, with disputes over whether it is IP theft or fair practice.Ep 90 - China isn’t ignoring AI risk, it’s rejecting America’s version of it · The Trivium China Podcast · Oct 6
- Discussion of how frontier labs use decompiled game data and RL environments to improve future models.AI’s Economic Impact, Warren Buffett Discovers YouTube, Mistral Launches “Le Chonk” | Diet TBPN · TBPN · Oct 6
- Discussion of whether LLMs have world models and how language models produce outputs, treated factually.This Dum Week 2026-10-04 · This Dum Week · Oct 5
- Inference throughput modeling depends on model architecture, context length and cache hit assumptions.Ep. 035 - Tech DD’s, Performance Projections, Benchmarks, Supply Chain, Investment Thesis (Consulting)| Abhilash Jain, Jordan Nanos · SemiAnalysis Weekly · Oct 5
- Discusses the unexplained effectiveness of scaling compute and data in language models.AGI Believers: Kevin Roose on the Race · ChinaTalk · Oct 8
- Experiments with Cloudflare's AI models and a humanizer tool, described factually.Cloudflare Edge Workers with Bob Fornal · RunAs Radio · Oct 7
- Discussion of model capability improvements lifting product utility is largely factual, without a clear investing angle.The Top 100 Consumer AI Apps: Who’s Actually Paying? · The a16z Show · Oct 5
- Technical explanation of cache hit rates, KV cache and inference serving economics.Ep. 036 - $200 Buys $12,000 of Opus Tokens, We Bought Every Plan (AI Cloud TCO) | Max Kan, Jordan Nanos, Andrew Megalaa · SemiAnalysis Weekly · Oct 7
- Discussion of AI, GPU compute and computational physics converging to enable higher-fidelity simulation.Podcast EP371: How Vinci is Revolutionizing Hardware Design with John Bruggeman · SemiWiki.com · Oct 6
Around it
Shows that covered it most
- MTS17
- TBPN4
- The Cognitive Revolution3
- Moonshots with Peter Diamandis2
- SemiAnalysis Weekly2
- The a16z Show2
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