After the failed OpenAI coup, AI development will likely move faster and be more dispersed and less controlled, accelerating what the coup was trying to prevent
Financial TimesBenedict Evans
Context & Ripple Effects
The failed attempt to change OpenAI's control came amid a broader argument over whether frontier-model development should be concentrated under tighter institutional oversight or spread more widely. Related coverage had already highlighted that the open-source ecosystem remained dependent on models controlled by large companies, complicating the idea that dispersion automatically means independence: open-source AI’s reliance on Big Tech models.
The failed takeover leaves OpenAI under its existing control structure, while the safety-oriented intervention it represented does not produce the intended restraint on development.
The immediate implication is a faster, more distributed and less centrally controlled AI-development race, increasing the difficulty of coordinating safeguards across participants.
Second-order effects
Commercial partners and rival labs gain reason to prioritize product delivery and talent retention over governance experiments that could interrupt development, as later reporting on commercial-priority pressure at OpenAI suggests.
More dispersed development broadens model access, but the dependence identified in the Big Tech-backed open-source stack means infrastructure and leading-model providers can still retain important leverage.
Third-order effects
If this pattern persists, frontier-AI safety will be harder to manage through the governance of any one lab; oversight will increasingly need to address model access, deployment and infrastructure across multiple actors.
The episode points to a durable tension between institutional controls designed to slow or constrain capability advances and market, partner and open-development pressures that diffuse those advances.
The trend: Frontier AI is moving toward a governance model in which control contests inside individual labs matter less than the wider commercial and distribution networks accelerating model development.
A techie delusion is thinking that enabling companies to replace workers with AI creates a post-scarcity economy. This isn't true because resources people need to live are still scarce — land, drinking water, food, healthcare, quality education, etc. …
It's becoming pretty common in the internet spaces I inhabit to say that AI is actually not that big of a deal and all of the fears about it are just free hype for AI companies. I feel like folks are having very different reactions to GPT fucking obliterating the Turing Test tha…
here's a thing I haven't seen anyone but @ryanhatesthis say and now I am also saying it at @wsj: The only thing you really need to know about the entire OpenAI drama just passed is that it's the dumbest thing ever. It's essentially a religious debate. …
This is such an important debate. A lot depends on whether a) there really is an existential danger to humanity from AGI and b) whether AGI is getting near. If, like me, you've been persuaded that the answer to both questions is probably yes, then any rational human should care…
I was a fan of Effective Altruism (almost taught a course on it at Harvard) together w other rational efforts (evidence-based medicine, data-driven policing, randomista econ). But it became cultish. Happy to donate to save the most lives in Africa, but not to pay techies to fret.…
Effective altruist now seems synonymous with AI doomer, but it wasn't always that way. My own experience & why I think a lot of it is bullshit AI doomerism now: I was an active member - went to multiple of their global summits and still mostly donate to causes that they...
When OpenAI was founded, its architects “recognised the commercial impulse could lead to disaster”, says prof Stuart Russel. “Now they've succumbed to the commercial impulse themselves.” OpenAI and the rift at the heart of Silicon Valley via @FT https://www.ft.com/...
“The engineers on the Apollo Program knew how far away the moon was and how much thrust the rocket had but we don't know how far away AGI is, nor how close OpenAI's large language models are, nor whether they can get there.” -@benedictevans🎯 🎯🎯 https://www.ft.com/... #ai #agi
“That seems to be the best explanation of what has happened at OpenAI: those who think we should slow down and be careful mounted a coup against those who think we should speed up and be careful”. And the “doomers” lost https://www.ft.com/...
This piece from @benedictevans in the @ft is very good. Doomers have a weak theory and sociology of power and societal change. https://www.ft.com/... [image]
Great piece by @benedictevans about the nativity of the @OpenAI board (and tech in many cases IMHO) when it comes to understanding the real nature of power, complexity and human systems and the management of societal impact by technology. https://www.ft.com/...