Jun 7, 2025 · 1h 57m · tbpn

Weekly Recap - Elon Vs Trump, Ukraine's Drone Attack, Cluely Update & OpenAI CRO

John Coogan · 35m spoken Mark Chen · 16m spoken Jordi Hays · 13m spoken Sholto Douglas · 11m spoken Kian Sadeghi · 6m spoken Delian Asparouhov · 5m spoken Roy Lee · 5m spoken Soren Monroe-Anderson · 3m spoken Connor Lovett · 3m spoken Donald Trump · 1m spoken Erik Prince · 1m spoken
0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

This TBPN weekly recap covers Ukraine's groundbreaking Operation Spiderweb drone offensive, the political and financial fallout of the Trump-Musk feud, and technical deep-dives into frontier AI reasoning models with OpenAI and Anthropic researchers.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 34.1% of the talking time here. How this is scored →

The hosts as informed peer 4.1 Guest teaching 4.4 Guest disagreement 0.6 The hosts pushing back 1.2
05100:0020:0040:001:00:001:20:001:40:002:58–9:01 · The hosts as informed peer 4/10 Operation Spiderweb: Soren Monroe-Anderson on Ukrainian Drone Warfare Coogan and Hays ask knowledgeable introductory questions about the evolution of Ukrainian drone tactics and leapfrogging Predator drones. Guest Soren Monroe-Anderson details the specifics of Operation Spiderweb and cellular network usage, educating the hosts without conflict.9:02–13:45 · The hosts as informed peer 4/10 Defense Tech Implications of Drone Warfare with Connor Lovett Connor Lovett provides venture capital and defense tech insights on autonomous systems and multi-layered defense. Coogan and Hays facilitate the conversation collaboratively.13:46–18:56 · The hosts as informed peer 0/10 Erik Prince on European Defense and Ukraine's Military Capabilities This clip features Erik Prince monologuing on European defense and Ukrainian tactical superiority compared to NATO forces. Host activity is not present in direct dialogue during this clip, requiring zeroed host scores.18:57–20:58 · The hosts as informed peer 0/10 Donald Trump Press Conference on Elon Musk and the Spending Bill This segment is a news clip playback of Donald Trump answering press questions regarding Elon Musk and the spending bill. Neither host nor guest dynamic applies.20:58–32:49 · The hosts as informed peer 5/10 Space and Business Fallout with Delian Asparouhov Delian Asparouhov unpacks the policy and space industry fallout between Trump and Musk, explaining the SLS program and Dragon spacecraft implications. Coogan contributes historical analogies about Gilded Age power structures and discusses Boeing's stock movements.32:50–46:53 · The hosts as informed peer 4/10 Cluely's Viral Growth Machine and Team Update with Roy Lee Roy Lee presents his ultra-aggressive viral marketing and compensation strategy for Cluely. Coogan and Hays lightly tease and question the product mechanics, asking if it resembles an MLM or if customers feel scammed.46:55–1:00:17 · The hosts as informed peer 6/10 Launching Nucleus Embryo: Genetic Optimization with Kian Sadeghi Coogan pushes Sadeghi on the regulatory classification, questioning where FDA oversight kicks in for embryo scoring and genetic predictions. Sadeghi defends the consumer agency model and demonstrates the website feature.1:00:18–1:28:53 · The hosts as informed peer 7/10 The Frontier of AI Reasoning and Agents with OpenAI's Mark Chen Coogan and Hays engage in a deeply technical discussion with Mark Chen regarding RL scaling, pre-training limits, and AI agent architectures. Coogan draws sharp distinctions between move 37 innovation, benchmarks like IOI/ARC-AGI, and deep research vs operator.1:28:55–1:56:57 · The hosts as informed peer 7/10 Scaling Reinforcement Learning and Model Architectures with Sholto Douglas Coogan drills into the engineering constraints of RL, memory bandwidth vs FLOPs in custom silicon, context windows, and physical energy bounds on scaling laws. Douglas provides detailed explanations of RL reward hacking and TPU architecture.2:58–9:01 · Guest teaching 6/10 Operation Spiderweb: Soren Monroe-Anderson on Ukrainian Drone Warfare Coogan and Hays ask knowledgeable introductory questions about the evolution of Ukrainian drone tactics and leapfrogging Predator drones. Guest Soren Monroe-Anderson details the specifics of Operation Spiderweb and cellular network usage, educating the hosts without conflict.9:02–13:45 · Guest teaching 5/10 Defense Tech Implications of Drone Warfare with Connor Lovett Connor Lovett provides venture capital and defense tech insights on autonomous systems and multi-layered defense. Coogan and Hays facilitate the conversation collaboratively.13:46–18:56 · Guest teaching 7/10 Erik Prince on European Defense and Ukraine's Military Capabilities This clip features Erik Prince monologuing on European defense and Ukrainian tactical superiority compared to NATO forces. Host activity is not present in direct dialogue during this clip, requiring zeroed host scores.18:57–20:58 · Guest teaching 0/10 Donald Trump Press Conference on Elon Musk and the Spending Bill This segment is a news clip playback of Donald Trump answering press questions regarding Elon Musk and the spending bill. Neither host nor guest dynamic applies.20:58–32:49 · Guest teaching 5/10 Space and Business Fallout with Delian Asparouhov Delian Asparouhov unpacks the policy and space industry fallout between Trump and Musk, explaining the SLS program and Dragon spacecraft implications. Coogan contributes historical analogies about Gilded Age power structures and discusses Boeing's stock movements.32:50–46:53 · Guest teaching 4/10 Cluely's Viral Growth Machine and Team Update with Roy Lee Roy Lee presents his ultra-aggressive viral marketing and compensation strategy for Cluely. Coogan and Hays lightly tease and question the product mechanics, asking if it resembles an MLM or if customers feel scammed.46:55–1:00:17 · Guest teaching 4/10 Launching Nucleus Embryo: Genetic Optimization with Kian Sadeghi Coogan pushes Sadeghi on the regulatory classification, questioning where FDA oversight kicks in for embryo scoring and genetic predictions. Sadeghi defends the consumer agency model and demonstrates the website feature.1:00:18–1:28:53 · Guest teaching 4/10 The Frontier of AI Reasoning and Agents with OpenAI's Mark Chen Coogan and Hays engage in a deeply technical discussion with Mark Chen regarding RL scaling, pre-training limits, and AI agent architectures. Coogan draws sharp distinctions between move 37 innovation, benchmarks like IOI/ARC-AGI, and deep research vs operator.1:28:55–1:56:57 · Guest teaching 5/10 Scaling Reinforcement Learning and Model Architectures with Sholto Douglas Coogan drills into the engineering constraints of RL, memory bandwidth vs FLOPs in custom silicon, context windows, and physical energy bounds on scaling laws. Douglas provides detailed explanations of RL reward hacking and TPU architecture.2:58–9:01 · Guest disagreement 0/10 Operation Spiderweb: Soren Monroe-Anderson on Ukrainian Drone Warfare Coogan and Hays ask knowledgeable introductory questions about the evolution of Ukrainian drone tactics and leapfrogging Predator drones. Guest Soren Monroe-Anderson details the specifics of Operation Spiderweb and cellular network usage, educating the hosts without conflict.9:02–13:45 · Guest disagreement 0/10 Defense Tech Implications of Drone Warfare with Connor Lovett Connor Lovett provides venture capital and defense tech insights on autonomous systems and multi-layered defense. Coogan and Hays facilitate the conversation collaboratively.13:46–18:56 · Guest disagreement 0/10 Erik Prince on European Defense and Ukraine's Military Capabilities This clip features Erik Prince monologuing on European defense and Ukrainian tactical superiority compared to NATO forces. Host activity is not present in direct dialogue during this clip, requiring zeroed host scores.18:57–20:58 · Guest disagreement 0/10 Donald Trump Press Conference on Elon Musk and the Spending Bill This segment is a news clip playback of Donald Trump answering press questions regarding Elon Musk and the spending bill. Neither host nor guest dynamic applies.20:58–32:49 · Guest disagreement 0/10 Space and Business Fallout with Delian Asparouhov Delian Asparouhov unpacks the policy and space industry fallout between Trump and Musk, explaining the SLS program and Dragon spacecraft implications. Coogan contributes historical analogies about Gilded Age power structures and discusses Boeing's stock movements.32:50–46:53 · Guest disagreement 2/10 Cluely's Viral Growth Machine and Team Update with Roy Lee Roy Lee presents his ultra-aggressive viral marketing and compensation strategy for Cluely. Coogan and Hays lightly tease and question the product mechanics, asking if it resembles an MLM or if customers feel scammed.46:55–1:00:17 · Guest disagreement 1/10 Launching Nucleus Embryo: Genetic Optimization with Kian Sadeghi Coogan pushes Sadeghi on the regulatory classification, questioning where FDA oversight kicks in for embryo scoring and genetic predictions. Sadeghi defends the consumer agency model and demonstrates the website feature.1:00:18–1:28:53 · Guest disagreement 1/10 The Frontier of AI Reasoning and Agents with OpenAI's Mark Chen Coogan and Hays engage in a deeply technical discussion with Mark Chen regarding RL scaling, pre-training limits, and AI agent architectures. Coogan draws sharp distinctions between move 37 innovation, benchmarks like IOI/ARC-AGI, and deep research vs operator.1:28:55–1:56:57 · Guest disagreement 1/10 Scaling Reinforcement Learning and Model Architectures with Sholto Douglas Coogan drills into the engineering constraints of RL, memory bandwidth vs FLOPs in custom silicon, context windows, and physical energy bounds on scaling laws. Douglas provides detailed explanations of RL reward hacking and TPU architecture.2:58–9:01 · The hosts pushing back 0/10 Operation Spiderweb: Soren Monroe-Anderson on Ukrainian Drone Warfare Coogan and Hays ask knowledgeable introductory questions about the evolution of Ukrainian drone tactics and leapfrogging Predator drones. Guest Soren Monroe-Anderson details the specifics of Operation Spiderweb and cellular network usage, educating the hosts without conflict.9:02–13:45 · The hosts pushing back 0/10 Defense Tech Implications of Drone Warfare with Connor Lovett Connor Lovett provides venture capital and defense tech insights on autonomous systems and multi-layered defense. Coogan and Hays facilitate the conversation collaboratively.13:46–18:56 · The hosts pushing back 0/10 Erik Prince on European Defense and Ukraine's Military Capabilities This clip features Erik Prince monologuing on European defense and Ukrainian tactical superiority compared to NATO forces. Host activity is not present in direct dialogue during this clip, requiring zeroed host scores.18:57–20:58 · The hosts pushing back 0/10 Donald Trump Press Conference on Elon Musk and the Spending Bill This segment is a news clip playback of Donald Trump answering press questions regarding Elon Musk and the spending bill. Neither host nor guest dynamic applies.20:58–32:49 · The hosts pushing back 1/10 Space and Business Fallout with Delian Asparouhov Delian Asparouhov unpacks the policy and space industry fallout between Trump and Musk, explaining the SLS program and Dragon spacecraft implications. Coogan contributes historical analogies about Gilded Age power structures and discusses Boeing's stock movements.32:50–46:53 · The hosts pushing back 2/10 Cluely's Viral Growth Machine and Team Update with Roy Lee Roy Lee presents his ultra-aggressive viral marketing and compensation strategy for Cluely. Coogan and Hays lightly tease and question the product mechanics, asking if it resembles an MLM or if customers feel scammed.46:55–1:00:17 · The hosts pushing back 3/10 Launching Nucleus Embryo: Genetic Optimization with Kian Sadeghi Coogan pushes Sadeghi on the regulatory classification, questioning where FDA oversight kicks in for embryo scoring and genetic predictions. Sadeghi defends the consumer agency model and demonstrates the website feature.1:00:18–1:28:53 · The hosts pushing back 3/10 The Frontier of AI Reasoning and Agents with OpenAI's Mark Chen Coogan and Hays engage in a deeply technical discussion with Mark Chen regarding RL scaling, pre-training limits, and AI agent architectures. Coogan draws sharp distinctions between move 37 innovation, benchmarks like IOI/ARC-AGI, and deep research vs operator.1:28:55–1:56:57 · The hosts pushing back 2/10 Scaling Reinforcement Learning and Model Architectures with Sholto Douglas Coogan drills into the engineering constraints of RL, memory bandwidth vs FLOPs in custom silicon, context windows, and physical energy bounds on scaling laws. Douglas provides detailed explanations of RL reward hacking and TPU architecture.

speaking balance: gold is the hosts, purple is the guest (3 minute bins)

0:00 · the hosts 74.8% · guest 25.2%0:00 · the hosts 74.8% · guest 25.2%3:00 · the hosts 23% · guest 77%3:00 · the hosts 23% · guest 77%6:00 · the hosts 0% · guest 100%6:00 · the hosts 0% · guest 100%9:00 · the hosts 13% · guest 87%9:00 · the hosts 13% · guest 87%12:00 · the hosts 0% · guest 100%12:00 · the hosts 0% · guest 100%15:00 · the hosts 20.1% · guest 79.9%15:00 · the hosts 20.1% · guest 79.9%18:00 · the hosts 4.5% · guest 95.5%18:00 · the hosts 4.5% · guest 95.5%21:00 · the hosts 27.3% · guest 72.7%21:00 · the hosts 27.3% · guest 72.7%24:00 · the hosts 26.8% · guest 73.2%24:00 · the hosts 26.8% · guest 73.2%27:00 · the hosts 31% · guest 69%27:00 · the hosts 31% · guest 69%30:00 · the hosts 24.7% · guest 75.3%30:00 · the hosts 24.7% · guest 75.3%33:00 · the hosts 49.8% · guest 50.2%33:00 · the hosts 49.8% · guest 50.2%36:00 · the hosts 24% · guest 76%36:00 · the hosts 24% · guest 76%39:00 · the hosts 39.6% · guest 60.4%39:00 · the hosts 39.6% · guest 60.4%42:00 · the hosts 25.6% · guest 74.4%42:00 · the hosts 25.6% · guest 74.4%45:00 · the hosts 23.4% · guest 76.6%45:00 · the hosts 23.4% · guest 76.6%48:00 · the hosts 17.1% · guest 82.9%48:00 · the hosts 17.1% · guest 82.9%51:00 · the hosts 35.4% · guest 64.6%51:00 · the hosts 35.4% · guest 64.6%54:00 · the hosts 46% · guest 54%54:00 · the hosts 46% · guest 54%57:00 · the hosts 56.2% · guest 43.8%57:00 · the hosts 56.2% · guest 43.8%1:00:00 · the hosts 45% · guest 55%1:00:00 · the hosts 45% · guest 55%1:03:00 · the hosts 49.2% · guest 50.8%1:03:00 · the hosts 49.2% · guest 50.8%1:06:00 · the hosts 17.5% · guest 82.5%1:06:00 · the hosts 17.5% · guest 82.5%1:09:00 · the hosts 23.8% · guest 76.2%1:09:00 · the hosts 23.8% · guest 76.2%1:12:00 · the hosts 47.4% · guest 52.6%1:12:00 · the hosts 47.4% · guest 52.6%1:15:00 · the hosts 36.7% · guest 63.3%1:15:00 · the hosts 36.7% · guest 63.3%1:18:00 · the hosts 16.7% · guest 83.3%1:18:00 · the hosts 16.7% · guest 83.3%1:21:00 · the hosts 16.7% · guest 83.3%1:21:00 · the hosts 16.7% · guest 83.3%1:24:00 · the hosts 28.6% · guest 71.4%1:24:00 · the hosts 28.6% · guest 71.4%1:27:00 · the hosts 48.9% · guest 51.1%1:27:00 · the hosts 48.9% · guest 51.1%1:30:00 · the hosts 41.6% · guest 58.4%1:30:00 · the hosts 41.6% · guest 58.4%1:33:00 · the hosts 44.6% · guest 55.4%1:33:00 · the hosts 44.6% · guest 55.4%1:36:00 · the hosts 69.5% · guest 30.5%1:36:00 · the hosts 69.5% · guest 30.5%1:39:00 · the hosts 62.8% · guest 37.2%1:39:00 · the hosts 62.8% · guest 37.2%1:42:00 · the hosts 51.3% · guest 48.7%1:42:00 · the hosts 51.3% · guest 48.7%1:45:00 · the hosts 67.2% · guest 32.8%1:45:00 · the hosts 67.2% · guest 32.8%1:48:00 · the hosts 52.2% · guest 47.8%1:48:00 · the hosts 52.2% · guest 47.8%1:51:00 · the hosts 29.4% · guest 70.6%1:51:00 · the hosts 29.4% · guest 70.6%1:54:00 · the hosts 19.9% · guest 80.1%1:54:00 · the hosts 19.9% · guest 80.1%1:57:00 · the hosts 0% · guest 0%1:57:00 · the hosts 0% · guest 0%
Sharpest disagreement ▶ 37:15 Roy Lee rejects product value in favor of pure viral attention

Roy Lee emphatically dismisses the premise of building traditional product value, arguing crudely that in 2025 product does not matter compared to sheer virality and eyeball capture.

Hardest push from the hosts ▶ 55:42 John Coogan challenges Nucleus on potential FDA regulation

Coogan refuses to accept Sadeghi's dismissive framing that trait analysis doesn't belong in medical regulation, pointing out democratic oversight and the breadth of FDA jurisdiction.

Biggest teaching moment ▶ 6:10 Soren Monroe-Anderson explains cellular radio control of FPV drones

Monroe-Anderson educates the hosts on how Ukrainian operatives ran short-range FPV drones 4000 km deep inside Russia using cellular networks instead of local RF.

The host holds their own ▶ 1:16:16 John Coogan contextualizes AlphaGo's Move 37 with modern LLMs

Coogan showcases his deep understanding of AI history by contrasting genuine generative innovation against current viral LLM mimicry using the iconic Move 37 frame.

the scores for every segment, with the reasoning behind each
ChapterTopicThe hosts as informed peerGuest teachingGuest disagreementThe hosts pushing backWhy
Operation Spiderweb: Soren Monroe-Anderson on Ukrainian Drone Warfare 4600 Coogan and Hays ask knowledgeable introductory questions about the evolution of Ukrainian drone tactics and leapfrogging Predator drones. Guest Soren Monroe-Anderson details the specifics of Operation Spiderweb and cellular network usage, educating the hosts without conflict.
Defense Tech Implications of Drone Warfare with Connor Lovett 4500 Connor Lovett provides venture capital and defense tech insights on autonomous systems and multi-layered defense. Coogan and Hays facilitate the conversation collaboratively.
Erik Prince on European Defense and Ukraine's Military Capabilities 0700 This clip features Erik Prince monologuing on European defense and Ukrainian tactical superiority compared to NATO forces. Host activity is not present in direct dialogue during this clip, requiring zeroed host scores.
Donald Trump Press Conference on Elon Musk and the Spending Bill 0000 This segment is a news clip playback of Donald Trump answering press questions regarding Elon Musk and the spending bill. Neither host nor guest dynamic applies.
Space and Business Fallout with Delian Asparouhov 5501 Delian Asparouhov unpacks the policy and space industry fallout between Trump and Musk, explaining the SLS program and Dragon spacecraft implications. Coogan contributes historical analogies about Gilded Age power structures and discusses Boeing's stock movements.
Cluely's Viral Growth Machine and Team Update with Roy Lee 4422 Roy Lee presents his ultra-aggressive viral marketing and compensation strategy for Cluely. Coogan and Hays lightly tease and question the product mechanics, asking if it resembles an MLM or if customers feel scammed.
Launching Nucleus Embryo: Genetic Optimization with Kian Sadeghi 6413 Coogan pushes Sadeghi on the regulatory classification, questioning where FDA oversight kicks in for embryo scoring and genetic predictions. Sadeghi defends the consumer agency model and demonstrates the website feature.
The Frontier of AI Reasoning and Agents with OpenAI's Mark Chen 7413 Coogan and Hays engage in a deeply technical discussion with Mark Chen regarding RL scaling, pre-training limits, and AI agent architectures. Coogan draws sharp distinctions between move 37 innovation, benchmarks like IOI/ARC-AGI, and deep research vs operator.
Scaling Reinforcement Learning and Model Architectures with Sholto Douglas 7512 Coogan drills into the engineering constraints of RL, memory bandwidth vs FLOPs in custom silicon, context windows, and physical energy bounds on scaling laws. Douglas provides detailed explanations of RL reward hacking and TPU architecture.

Statements from this episode (23)

Assertion Contradicted
Monroe-Anderson: Ukraine Will Produce 4.5M FPV Drones in 2025
“It's estimated Ukraine is going to produce four and a half million FPV drones this year.”
Soren Monroe-Anderson Jun 7, 2025 ▶ 5:14
Assertion Supported
Monroe-Anderson: Operation Spiderweb Drones Operated on Cellular Networks
“And the drones were actually operating on cellular. They were not operating on Local, like the normal low latency local radios you use for FPVs typically.”
Soren Monroe-Anderson Jun 7, 2025 ▶ 7:11
Opinion
Monroe-Anderson: US Has No Defenses Against Covert Drone Attacks
“I mean, being blatant, the US does not have any defenses in place that would stop this from happening. We already know we, there's already news stories about drones that are flying over our air force bases, and we can't do anything about it.”
Soren Monroe-Anderson Jun 7, 2025 ▶ 8:25
Assertion Not checkable as stated
Lovett: Shutting down cell networks impairs Russian military coordination on WhatsApp
“This is like a dirty little secret that nobody talks about. You know, yes, you have your military systems that are protected and all that, But a lot of coordination is happening through WhatsApp, a lot of coordination on, and so all of a sudden you turn off th…”
Connor Lovett Jun 7, 2025 ▶ 12:53
Opinion
Prince: Ukraine possesses the most capable military in Europe
“The most capable military in Europe right now is the Ukrainian military.”
Erik Prince Jun 7, 2025 ▶ 13:47
Opinion
Prince: Ukraine's EW and close air support surpass US military
“Their ability to fight against and to even conduct electronic warfare and even close air support in this environment is is leaps and bounds ahead of even what the US military is.”
Erik Prince Jun 7, 2025 ▶ 13:59
Prediction Not checkable as stated
Prince: Blitzkrieg maneuver warfare impossible in Ukraine without new weapons
“You're not going to see any kind of blitzkrieg, Hans Guderian maneuver warfare there until some significantly different weapon systems come along.”
Erik Prince Jun 7, 2025 ▶ 15:05
Assertion Not checkable as stated
Trump: Musk Opposed Spending Bill Because EV Mandates Were Cut
“All of a sudden he had a problem, and he only developed the problem when he found out that we're going to have to cut the EV mandate, because that's billions and billions of dollars, and it really is unfair.”
Donald Trump Jun 7, 2025 ▶ 19:59
Prediction Held up
Trump: Musk will criticize me personally next
“He hasn't said bad about me personally, but I'm sure that'll be next.”
Donald Trump Jun 7, 2025 ▶ 20:21
Assertion Supported
Asparouhov: Senate Package Restores NASA Funding for SLS Rocket
“It was released that the Senate reconciliation package re-added budget back into NASA, largely for the SLS program, which was basically the program that, you know, sort of Jared and Elon We're, you know, sort of largely advocating to, you know, sort of complet…”
Delian Asparouhov Jun 7, 2025 ▶ 22:33
Prediction Open · timeframe Jun 2027
Asparouhov: Starlink Revenue Will Pass NASA's Budget Within Two Years
“Starlink's total top line revenue is going to be passing the NASA budget in the next year or two.”
Delian Asparouhov Jun 7, 2025 ▶ 24:53
Insight
Roy Lee: No Company Ever Dies From Founder Controversy
“There's no company that ever died from a founder being too controversial. You got deal fucking infiltrating with genuine spies. And they're still doing fine, bro. Worker, 17 guys. They're still kicking. Like, no company ever dies from being too controversial. …”
Roy Lee Jun 7, 2025 ▶ 42:43
Assertion Supported
Sadeghi: Height has the best polygenic predictor in genomics today
“Even though the best predictor we have today, actually in the world, the best polygenic predictor is for height.”
Kian Sadeghi Jun 7, 2025 ▶ 50:22
Prediction Not checkable as stated
Chen: 2025 will be the year of autonomous AI agents
“We see 25 as this year of agents, right? We think of it as a year where models are going to do a lot more autonomous work. You can let them Kind of be unsupervised for much longer periods of time.”
Mark Chen Jun 7, 2025 ▶ 1:05:08
Assertion Supported
Chen: OpenAI reached 3 million paying business users
“We hit a big milestone. We got I think three million paying business users fairly recently.”
Mark Chen Jun 7, 2025 ▶ 1:07:26
Assertion Supported
Chen: Frontier AI models consistently score over 90% on AIME
“I think one clear example here is the Amy, like probably the hardest auto gradable, like human math eval, at least in the US. And yeah, the models are consistently getting like 90 plus percent on these.”
Mark Chen Jun 7, 2025 ▶ 1:08:50
Insight
Chen: AI reasoning only emerges at large scale requiring massive compute
“When you look at reasoning you just don't see that happen at small scale, right? There's like a certain scale at which it starts becoming signal bearing and that requires you to have resources, right?”
Mark Chen Jun 7, 2025 ▶ 1:22:48
Insight
Chen: AI reasoning is the key mechanism to make agents reliable
“And I think the reason why we care so much about reasoning is because I think that's the path that we get reliable agents through.”
Mark Chen Jun 7, 2025 ▶ 1:26:43
Insight
Mark Chen: Compute Can Scale Heavily into RL Given Right Levers
“I think, like, if you find the right levers, you can really pump a lot of compute into RL as well as pre-training.”
Mark Chen Jun 7, 2025 ▶ 1:28:16
Assertion Not checkable as stated
Douglas: 10x RL compute increases still deliver linear performance gains
“We're still seeing these huge gains where you go, you know, 10 X compute increase in RL. We still getting like very distinct linear gains for that.”
Sholto Douglas Jun 7, 2025 ▶ 1:30:12
Prediction Not checkable as stated
Douglas: Generalist models will obsolete specialized fine-tuned models
“I really do think that similar to how we saw with large pre-trained models before with small fine-tuned models made it like, had gains over the sort of GPT-II era, but then were obsoleted by GPT-IV being generally good at everything. I think, to be honest, you…”
Sholto Douglas Jun 7, 2025 ▶ 1:34:23
Prediction Not checkable as stated
Douglas: AI will use long-horizon reward signals within two years
“We've got very little rewards right now, but pretty quickly over the next year or two, you're going to start to see much more meaningful and long horizon rewards.”
Sholto Douglas Jun 7, 2025 ▶ 1:35:46
Disclosure
Douglas: Frontier Labs Avoid RL Training Directly on ARC-AGI
“And I mean, I think if you are old on Arc AGI, then it would, you'd probably get superhuman at it pretty fast. But I think we're all trying not to RL on it so that it functions as like an interesting held out.”
Sholto Douglas Jun 7, 2025 ▶ 1:55:15
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 500 episodes transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.