TBPN
TBPN

Fruitfly Hard Takeoff, Washington on AI Risk, đť•Ź Timeline Reactions | Thijs Simonian, Alex Heath & Guy Oseary, Mitesh Agrawal

September 11, 2026

AI Summary

5 min read

The AI Doomer Playbook Is More Nuanced Than You Think

The most concrete policy proposal to come out of the AI risk community isn't about shutting everything down. AI 2040, the sequel to the widely circulated AI 2027 document, explicitly wants to reach superintelligence—just by 2040 instead of 2028. The plan is not to stop AI but to slow its acceleration to a pace where society can keep up, and it contains a remarkably specific set of technical enforcement mechanisms that show how far the doomer camp has moved from vague warnings into detailed regulatory engineering.

The Core Mechanism: Compute Caps as the Main Valve

The central insight of the AI 2040 proposal is that the most effective way to control AI capability improvements is to control compute. The argument rests on a consensus that scale is a prerequisite for frontier models—that even if someone discovers a more algorithmically efficient path to AGI, they still need massive clusters of hardware. This makes compute a natural choke point.

Continue reading the full summary in the app — free to try.

Read Full Summary →

Free • No credit card required

Listen to Audio Summary Open in App

What you'll learn

  • 1 (01:44) **Opening: The AI Doomer Policy Conversation** - The host sets up the episode’s core question: what are AI doomers actually proposing, and how concretely are they trying to slow down development?
  • 2 (05:32) **The AI 2040 Proposal: A Concrete Slow-Down Plan** - The host introduces the AI 2040 proposal as the opposite of a grassroots shutdown: it aims to harden data centers and slow the race to superintelligence, not stop current models.
  • 3 (08:56) **Hardware-Level Enforcement: Ferret Cages and Capped Pipes** - The host details the extreme physical security measures proposed for new R&D data centers, designed to make model theft and secret training nearly impossible.
  • 4 (12:47) **The Core Valve: Compute Caps and Algorithmic Leakage** - The host explains that the most effective lever for slowing AI is controlling hardware, but this creates an incentive for secret algorithmic breakthroughs.
  • 5 (17:52) **The Balancing Act: Slowing Down vs. Losing the Good Ending** - The host weighs the costs and benefits of a slow-down, acknowledging the potential for regulatory capture and lost economic gains, but concluding the current models are already very useful.
  • 6 (24:12) **Bernie Sanders’ “Corporate Death Penalty” Bill** - The host pivots to Senator Bernie Sanders’ proposed legislation, which introduces far more aggressive penalties for AI developers.
  • 7 (27:37) **Industry Reactions: The Application Layer vs. The Frontier** - The host surveys reactions from different parts of the AI ecosystem, showing a split between those building on top of models and those building the models themselves.

+ Full timestamped outline available in the app

Guests on this episode

Alex Heath

Show Notes

  • (01:54) - Washington on AI Risk
  • (23:07) - đť•Ź Timeline Reactions
  • (34:23) - Fruitfly Hard Takeoff
  • (41:07) - Thijs Simonian discusses his exploratory robotics work as an OpenAI intern, including connecting Codex to an inexpensive, open-source robotic arm that can plan, paint, monitor its progress, and improve through iteration. He highlights the potential for affordable robots to handle everyday tasks such as sorting mail or cooking, while noting current limitations in speed, cost, image processing, and safety.
  • (55:31) - Alex Heath & Guy Oseary. Alex Heath discusses his transition from veteran technology journalist and Sources podcast host to venture investor at Sound Ventures. He explains how years of interviewing leading founders sharpened his instinct for evaluating companies and outlines his continued plans for insightful podcasts and newsletters covering technology, AI agents, and entrepreneurship. Guy Oseary is a music executive, talent manager, and technology investor who manages Madonna and the Red Hot Chili Peppers and previously managed U2. He is also a co-founder and general partner of Sound Ventures, the venture firm he started with Ashton Kutcher, and previously served as chairman of Maverick Records, where he helped build a label that sold more than 100 million albums.
  • (01:27:29) - Mitesh Agarwal discusses Positron AI’s development of memory-focused inference chips designed to offer scalable, cost-efficient alternatives within Nvidia’s ecosystem. He covers rapid semiconductor development, strong demand from hyperscalers and AI labs, manufacturing and deployment challenges, and the importance of open-source software and customer-specific optimization.


TBPN is made possible by:

Ramp - https://ramp.com

Public - https://public.com

Cisco - https://www.cisco.com

Console - https://www.console.com

CrowdStrike - https://www.crowdstrike.com

Figma - https://www.figma.com

MongoDB - https://www.mongodb.com

NYSE - https://www.nyse.com

Railway - https://railway.com

Shopify - https://www.shopify.com

Codex - http://openAI.com/codex


Follow TBPN: 

https://TBPN.com

https://x.com/tbpn

https://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231

https://podcasts.apple.com/us/podcast/tbpn/id1772360235

https://www.youtube.com/@TBPNLive

TBPN

More from this podcast

TBPN →