Prof G MarketsWhy OpenAI And Anthropic Are Pumping The Brakes
15 Sept 2026 · 28 minDespite recent public calls for an AI slowdown from major labs, the industry's real intention is not to halt development but to integrate more compute for safety and monitoring, indicating a continued, rather than curtailed, expansion of AI capabilities and infrastructure.
Summary
The podcast critically examines the recent public statements by OpenAI and Anthropic regarding the need to "pace the frontier" of AI development, exploring whether these calls signify a genuine shift towards caution, a strategic move for regulatory capture, or a pragmatic response to the challenges of scaling advanced AI. The discussion delves into the implications for the AI industry's growth, the perceived risks of superintelligence, and the current state of AI lab profitability, offering a more grounded, less alarmist perspective than the popular debate.
Interesting ideas
AI 'Slowdown' is a Reallocation of Compute, Not a Halt. The guest argues that calls from OpenAI and Anthropic to "pace the frontier" of AI development aren't about stopping progress but reallocating resources. They intend to dedicate significantly more compute, potentially 20% or more, to safety, monitoring, and alignment research. This means continued advancement in AI capabilities, but with increased efforts to ensure responsible integration, which could paradoxically lead to even more aggressive compute build-outs rather than a slowdown.
AI Risks are More 'Annoying' Swarms Than 'Apocalyptic' Catastrophes. While acknowledging tail risks, the guest dismisses the most extreme 'AI apocalypse' scenarios. Instead, the primary immediate risk from advanced LLMs is the potential for 'swarms of AI agents' to cause widespread 'annoyances' and 'pain' on the internet, similar to the 'Hugging Face incident.' This perspective suggests that the world will need to harden against these systems, but the benefits of AI are likely to outweigh these short-term disruptions.
Public Statements on AI Safety May Be Strategic Optics. The guest speculates that the public pronouncements on AI safety by lab leaders like Sam Altman and Dario Amadei could serve an optical purpose. While acknowledging genuine concern, these statements also position the companies as responsible actors. In a duopoly, neither company can afford to appear less cautious than the other, creating a game theory dynamic where public safety rhetoric offers reputational benefits and potentially influences regulatory direction without necessarily slowing internal development.
Anthropic's Profitability Claims Are Highly Skeptical. Reports of Anthropic's profitability are met with skepticism due to reliance on 'adjusted operating basis' accounting. This method excludes major expenses like revenue-sharing agreements and model training costs, which are the largest expenses for AI labs. Without full transparency in their S1 filing, the claim of profitability is considered 'meaningless,' casting doubt on the viability of the current AI business model for frontier labs, which face enormous operational losses like OpenAI's $20 billion.
Questions worth thinking about
If 'pacing the frontier' means more compute for safety, what are the implications for smaller AI labs or open-source initiatives that may not have the resources for such extensive safety allocations?
If the most significant near-term risks are 'annoyances' rather than 'catastrophes,' how should society and policymakers prioritize resources between hardening against these 'annoyances' and preparing for more extreme, albeit less likely, tail risks?
Written by AI from the episode's full transcript. It can get things wrong: if something here looks off, tell us at hello@podpings.com.