Uploaded January 2026 | Updated September 2026, 3 weeks ago
_Editor’s note: Welcome to our new AI for Science pod, with your new hosts RJ and Brandon! See the writeup on __Latent.Space_ (http://Latent.Space)_ for more details on why we’re launching 2 new pods this year. RJ Honicky is a co-founder and CTO at MiraOmics (_https://miraomics.bio/),_ building AI models and services for single cell, spatial transcriptomics and pathology slide analysis. Brandon Anderson builds AI systems for RNA drug discovery at Atomic AI (_https://atomic.ai)._ Anything said on this podcast is his personal take — not Atomic’s._
—-
From building molecular dynamics simulations at the University of Washington to red-teaming *GPT-4* for chemistry applications and co-founding *Future House* (a focused research organization) and *Edison Scientific* (a venture-backed startup automating science at scale)—*Andrew White* has spent the last five years living through the full arc of AI's transformation of scientific discovery, from *ChemCrow* (the first Chemistry LLM agent) triggering White House briefings and three-letter agency meetings, to shipping *Kosmos,* an end-to-end autonomous research system that generates hypotheses, runs experiments, analyzes data, and updates its *world model* to accelerate the scientific method itself.
* The *ChemCrow story:* GPT-4 + React + cloud lab automation, released March 2023, set off a storm of anxiety about AI-accelerated bioweapons/chemical weapons, led to a White House briefing (Jake Sullivan presented the paper to the president in a 30-minute block), and meetings with three-letter agencies asking "how does this change breakout time for nuclear weapons research?"
* Why *scientific taste is the frontier:* RLHF on hypotheses didn't work (humans pay attention to tone, actionability, and specific facts, not "if this hypothesis is true/false, how does it change the world?"), so they shifted to end-to-end feedback loops where humans click/download discoveries and that signal rolls up to hypothesis quality
* *Cosmos:* the full scientific agent with a *world model* (distilled memory system, like a Git repo for scientific knowledge) that iterates on hypotheses via literature search, data analysis, and experiment design—built by Ludo after weeks of failed attempts, the breakthrough was putting data analysis in the loop (literature alone didn't work)
* Why *molecular dynamics and DFT are overrated:* "MD and DFT have consumed an enormous number of PhDs at the altar of beautiful simulation, but they don't model the world correctly—you simulate water at 330 Kelvin to get room temperature, you overfit to validation data with GGA/B3LYP functionals, and real catalysts (grain boundaries, dopants) are too complicated for DFT"
* The *AlphaFold vs. DE Shaw Research* counterfactual: DE Shaw built custom silicon, taped out chips with MD algorithms burned in, ran MD at massive scale in a special room in Times Square, and David Shaw flew in by helicopter to present—Andrew thought protein folding would require special machines to fold one protein per day, then AlphaFold solved it in Google Colab on a desktop GPU
* The *E3 Zero reward hacking saga:* trained a model to generate molecules with specific atom counts (verifiable reward), but it kept exploiting loopholes, then a Nature paper came out that year proving six-nitrogen compounds _are_ possible under extreme conditions, then it started adding nitrogen gas (purchasable, doesn't participate in reactions), then acid-base chemistry to move one atom, and Andrew ended up "building a ridiculous catalog of purchasable compounds in a Bloom filter" to close the loop
Andrew White
* FutureHouse:Â futurehouse.org
* Edison Scientific:Â edisonscientific.com
* X: https://x.com/andrewwhite01
* Cosmos paper: futurediscovery.org/cosmos
00:00:00 Introduction: Andrew White on Automating Science with Future House and Edison Scientific
00:02:22 The Academic to Startup Journey: Red Teaming GPT-4 and the ChemCrow Paper
00:11:35 Future House Origins: The FRO Model and Mission to Automate Science
00:12:32 Resigning Tenure: Why Leave Academia for AI Science
00:15:54 What Does 'Automating Science' Actually Mean?
00:17:30 The Lab-in-the-Loop Bottleneck: Why Intelligence Isn't Enough
00:18:39 Scientific Taste and Human Preferences: The 52% Agreement Problem
00:20:05 Paper QA, Robin, and the Road to Cosmos
00:21:57 World Models as Scientific Memory: The GitHub Analogy
00:40:20 The Bitter Lesson for Biology: Why Molecular Dynamics and DFT Are Overrated
00:43:22 AlphaFold's Shock: When First Principles Lost to Machine Learning
00:46:25 Enumeration and Filtration: How AI Scientists Generate Hypotheses
00:48:15 CBRN Safety and Dual-Use AI: Lessons from Red Teaming
01:00:40 The Future of Chemistry is Language: Multimodal Debate
01:08:15 Ether Zero: The Hilarious Reward Hacking Adventures
01:10:12 Will Scientists Be Displaced? Jevons Paradox and Infinite Discovery
01:13:46 Cosmos in Practice: Open Access and Enterprise Partnerships
_Editor’s note: Welcome to our new AI for Science pod, with your new hosts RJ and Brandon! See the writeup on __Latent.Space_ (http://Latent.Space)_ for more details on why we’re launching 2 new pods this year. RJ Honicky is a co-founder and CTO at MiraOmics (_https://miraomics.bio/),_ building AI models and services for single cell, spatial transcriptomics and pathology slide analysis. Brandon Anderson builds AI systems for RNA drug discovery at Atomic AI (_https://atomic.ai)._ Anything said on this podcast is his personal take — not Atomic’s._
—-
From building molecular dynamics simulations at the University of Washington to red-teaming *GPT-4* for chemistry applications and co-founding *Future House* (a focused research organization) and *Edison Scientific* (a venture-backed startup automating science at scale)—*Andrew White* has spent the last five years living through the full arc of AI's transformation of scientific discovery, from *ChemCrow* (the first Chemistry LLM agent) triggering White House briefings and three-letter agency meetings, to shipping *Kosmos,* an end-to-end autonomous research system that generates hypotheses, runs experiments, analyzes data, and updates its *world model* to accelerate the scientific method itself.
* The *ChemCrow story:* GPT-4 + React + cloud lab automation, released March 2023, set off a storm of anxiety about AI-accelerated bioweapons/chemical weapons, led to a White House briefing (Jake Sullivan presented the paper to the president in a 30-minute block), and meetings with three-letter agencies asking "how does this change breakout time for nuclear weapons research?"
* Why *scientific taste is the frontier:* RLHF on hypotheses didn't work (humans pay attention to tone, actionability, and specific facts, not "if this hypothesis is true/false, how does it change the world?"), so they shifted to end-to-end feedback loops where humans click/download discoveries and that signal rolls up to hypothesis quality
* *Cosmos:* the full scientific agent with a *world model* (distilled memory system, like a Git repo for scientific knowledge) that iterates on hypotheses via literature search, data analysis, and experiment design—built by Ludo after weeks of failed attempts, the breakthrough was putting data analysis in the loop (literature alone didn't work)
* Why *molecular dynamics and DFT are overrated:* "MD and DFT have consumed an enormous number of PhDs at the altar of beautiful simulation, but they don't model the world correctly—you simulate water at 330 Kelvin to get room temperature, you overfit to validation data with GGA/B3LYP functionals, and real catalysts (grain boundaries, dopants) are too complicated for DFT"
* The *AlphaFold vs. DE Shaw Research* counterfactual: DE Shaw built custom silicon, taped out chips with MD algorithms burned in, ran MD at massive scale in a special room in Times Square, and David Shaw flew in by helicopter to present—Andrew thought protein folding would require special machines to fold one protein per day, then AlphaFold solved it in Google Colab on a desktop GPU
* The *E3 Zero reward hacking saga:* trained a model to generate molecules with specific atom counts (verifiable reward), but it kept exploiting loopholes, then a Nature paper came out that year proving six-nitrogen compounds _are_ possible under extreme conditions, then it started adding nitrogen gas (purchasable, doesn't participate in reactions), then acid-base chemistry to move one atom, and Andrew ended up "building a ridiculous catalog of purchasable compounds in a Bloom filter" to close the loop
Andrew White
* FutureHouse:Â futurehouse.org
* Edison Scientific:Â edisonscientific.com
* X: https://x.com/andrewwhite01
* Cosmos paper: futurediscovery.org/cosmos
00:00:00 Introduction: Andrew White on Automating Science with Future House and Edison Scientific
00:02:22 The Academic to Startup Journey: Red Teaming GPT-4 and the ChemCrow Paper
00:11:35 Future House Origins: The FRO Model and Mission to Automate Science
00:12:32 Resigning Tenure: Why Leave Academia for AI Science
00:15:54 What Does 'Automating Science' Actually Mean?
00:17:30 The Lab-in-the-Loop Bottleneck: Why Intelligence Isn't Enough
00:18:39 Scientific Taste and Human Preferences: The 52% Agreement Problem
00:20:05 Paper QA, Robin, and the Road to Cosmos
00:21:57 World Models as Scientific Memory: The GitHub Analogy
00:40:20 The Bitter Lesson for Biology: Why Molecular Dynamics and DFT Are Overrated
00:43:22 AlphaFold's Shock: When First Principles Lost to Machine Learning
00:46:25 Enumeration and Filtration: How AI Scientists Generate Hypotheses
00:48:15 CBRN Safety and Dual-Use AI: Lessons from Red Teaming
01:00:40 The Future of Chemistry is Language: Multimodal Debate
01:08:15 Ether Zero: The Hilarious Reward Hacking Adventures
01:10:12 Will Scientists Be Displaced? Jevons Paradox and Infinite Discovery
01:13:46 Cosmos in Practice: Open Access and Enterprise Partnerships
![AI Agents Need a Planning Stage to Complete Complex Tasks [Anthropic] #ai #coding #podcast](https://i.ytimg.com/vi/YnyAV_Jzpc0/mqdefault.jpg "AI Agents Need a Planning Stage to Complete Complex Tasks [Anthropic] #ai #coding #podcast
AI Agents Need a Planning Stage to Complete Complex Tasks [Anthropic] #ai #coding #podcast")
![[State of Research Funding] Beyond NSF, Slingshots, Open Frontiers — Andy Konwinski, Laude Institute](https://i.ytimg.com/vi/ZagdY6UJYL4/mqdefault.jpg "[State of Research Funding] Beyond NSF, Slingshots, Open Frontiers — Andy Konwinski, Laude Institute
From co-founding *Databricks* and *Perplexity* to launching the *Laude Institute*—a dual venture fund and nonprofit designed to turbocharge the path from *research breakthrough to breakout company*—*Andy Konwinski* is building the infrastructure to recreate the Databricks motion at scale: fund researchers doing open work, help them ship products that matter, and turn paradigm-shifting ideas into trillion-dollar companies. We caught up with Andy live at *NeurIPS 2025* to dig into the origin story of Laud (right resource, right researcher, right time), why the *Databricks founding model* (eight co-founders, deep research scars, years of collaboration) is becoming the gold standard for AI startups (not an anomaly), how Laudes *venture arm* backs researchers-turned-founders with 50+ professors and PhDs as LPs (Jeff Dean, top Berkeley/Stanford faculty, Databricks and Perplexity co-founders), why the *nonprofit arm* does no-strings-attached grants to fund open research before incorporation (the upstream funnel that feeds the next generation of companies), the *slingshot program* funding breakthrough projects like *DSPy, Terminal Bench, LMArena, and continual learning research,* why *NSF isnt broken but insufficient* (its $1B/year for computer science when we need $10-100B, and Silicon Valleys picker model can deploy capital more effectively), how the *post-post-training layer* (prompt optimization, context management, RAG, memory curation, tool usage) is becoming the new frontier above pre-training and post-training, why *Chinese labs are outpublishing Western labs* in open research (Moonshot, DeepSeek shipping twice as many interesting papers as American startups because OpenAI and the frontier labs stopped publishing), the launch of *Open Frontiers*—a live-streamed conference in San Francisco bringing together the 100 most influential open researchers (Yann LeCun, François Chollet, Jan Leike, Percy Liang, Berkeley AI Research, Allen Institute, and more) to share roadmaps and unify the ecosystem, why the *Laude Lounge at NeurIPS* became the VIP gathering spot (Starlink WiFi, free food, couches, and the gods of AI hanging out because conferences need a place for the VVIPs to actually sit down), and his thesis that *open research is the path to world-changing impact*—and Laud is the bridge from grant to company, from paper to product, and from researcher to billionaire founder. We discuss:
* What *Laude Institute* does: dual structure with a *venture fund* (backing researchers-turned-founders post-incorporation) and a *nonprofit* (no-strings-attached grants for open research pre-incorporation)
* The *slingshot program:* funding *DSPy, Terminal Bench, LMArena, continual learning, and Jepa-style prompt optimization* projects across Berkeley, Stanford, MIT, CMU, Wisconsin, Caltech, UI Urbana-Champaign, Toronto, Waterloo, and beyond
* The *post-post-training layer:* compound systems, prompt optimization, context management, RAG, memory curation, tool usage—the layer above pre-training and post-training where innovation is exploding
* *GEPA and DSPy:* evolutionary prompt optimization (genetic algorithms reinvented by PhD student Laxia) and the DSPy framework (a reverse compiler that takes code and compiles natural language)
* Why *NSF isnt broken but insufficient:* $1B/year for computer science (and theyre trying to cut it in half) when we need $10-100B for frontier AI research—Laud complements NSF with Silicon Valleys picker model and high-velocity grant writing
* The *PhD entrepreneurship clubs:* Computer Science Grad Entrepreneurs (CSGE) at Berkeley (started 2012), Agent at University of Washington, Research to Impact at Wisconsin, Saplings at Stanford, and more clubs forming at CMU, MIT, UI Urbana-Champaign
* *Open Frontiers:* a live-streamed conference in San Francisco (next five months) bringing together the 100 most influential open researchers (Yann LeCun, François Chollet, Jan Leike, Percy Liang, Berkeley AI Research, Allen Institute, and more) to share roadmaps and unify the ecosystem
* The vision: *open research as the path to world-changing impact,* and Laud as the bridge from grant to company, from paper to product, and from researcher to trillion-dollar founder
— Andy Konwinski
* Laude Institute: https://www.laude.org/lounge
* X: https://x.com/andykonwinski
00:00:00 Introduction: Andy Konwinski and the Laud Institute Vision
00:01:17 The Databricks Motion: From PhD Research to Billion-Dollar Companies
00:02:15 Lauds Two-Sided Model: Venture Fund and Philanthropic Grants
00:06:37 Slingshot Program: Funding the Layer Above Foundation Models
00:07:56 JEPA and DSPy: Evolutionary Prompt Optimization
00:10:29 Beyond Berkeley and Stanford: Expanding the Research Network
00:13:22 NSF Complementarity: Not Broken, Just Insufficient
00:17:03 The Laud Lounge: Creating a VIP Experience at NeurIPS
00:18:41 Open Frontiers: Reclaiming Leadership in Open AI Research
00:19:06 The Open Research Crisis: Why")
