Get live statistics and analysis of Seán Ó hÉigeartaigh's profile on X / Twitter

It's not done yet. Hearing reports that the Nobel prize for literature will be going to the authors of "OpenAI's nonprofit governance structure" for outstanding contributions to creative fiction.

The year is 2025. The CEOs of two of the world's leading AI companies have (i) told the President of the United States of America that AGI will be developed in his presidency and (ii) told the world it will likely happen in 2026-27. 1/9

A small handful of Thiel business associates and a16z/Scale AI executives literally occupy every key AI position in USG, from which lofty position they tell us about regulatory capture. I love 2025, peak comedy.

I'm pretty confident we won't have AGI/country of geniuses in a datacenter within 2 years. I like ai-2027 as a piece of futures work, but I think too many people are treating it as a mainline scenario, rather than unlikely-but-not-impossible. I think this is resulting in too many people giving up on directions that might actually work (on realistic timelines) in favour of hail marys or flakier things. I also think it's contributing to things that will cause more harm than good, like leaning into the US-China AI race narrative (I think this will be the next clear example of Xrisk/EA folk hastening that which they aim to prevent - see e.g. OpenAI/Anthopic arc for previous).

Some "experts" claim that a single bipedal primate species designed all these wildly different modes of transport. The ridiculousness of this claim neatly illustrated the ridiculousness of the "AGI believers".

"Losing oversight and control of advanced AI systems, particularly Artificial General Intelligence (AGI), would be catastrophic. It must be avoided at all costs." Peter Kyle, UK Technology Secretary

Flew from Beijing Daixing for the first time today. It's a genuinely beautiful and thoughtfully designed building, and the effect on mood, and pleasantness of the airport experience, is noticeable. Photo is of the garden next to my departure gate. I sometimes feel we're being robbed in the ugliness of our infrastructure. The British managed it in the 19th Century (see the glory that is St Pancras) and the Chinese are managing it in the 21st Century, so why can't this be more common?

Let me say one thing that actually matters: if we believe their staff then what we're seeing, and that we're seeing eval results for, is not OpenAI's most capable model. They have more capable in-house. Without transparency requirements, this asymmetry will only grow, and will make meaningful awareness and governance increasingly ineffective + eventually impossible. The community needs to keep pushing for transparency requirements and testing of models in development and deployed internally.

AI Summit day 2 summary coming when I've slept and am less angry. What a shitshow.

Aside for the content (which I'm obviously sympathetic to), it's striking how much more skilled this three person outfit is at comms than a $500 billion company with infinity PR agencies at its disposal.

Recommend OpenAI subpoena these AGs to find out if they're really Elon Musk and three ferrets in a very large trenchcoat.

Up until 2025, what i would hear was "we can't afford to invest in safety because China won't invest in safety". Now that it's increasingly clear that the appetite for safety and safety-focused international governance is there for China, I fully expect to hear the opposite: if Chinese voices are calling for it, it can't be in 'our' interests to do it. My opinion: powerful AI interests with deep pockets don't want to be subject to governance, oversight, or safety requirements. Make them be.

Actually I think it's important to end the Summit on a positive note: now we can all finally give up the polite pretence that Mistral are a serious frontier AI player. Always a positive if you look hard enough.

The year is 2025. The CEOs of two of the world's leading AI companies have (i) told the President of the United States of America that AGI will be developed in his presidency and (ii) told the world it will likely happen in 2026-27. 1/9

I'm pretty confident we won't have AGI/country of geniuses in a datacenter within 2 years. I like ai-2027 as a piece of futures work, but I think too many people are treating it as a mainline scenario, rather than unlikely-but-not-impossible. I think this is resulting in too many people giving up on directions that might actually work (on realistic timelines) in favour of hail marys or flakier things. I also think it's contributing to things that will cause more harm than good, like leaning into the US-China AI race narrative (I think this will be the next clear example of Xrisk/EA folk hastening that which they aim to prevent - see e.g. OpenAI/Anthopic arc for previous).

A small handful of Thiel business associates and a16z/Scale AI executives literally occupy every key AI position in USG, from which lofty position they tell us about regulatory capture. I love 2025, peak comedy.

It's not done yet. Hearing reports that the Nobel prize for literature will be going to the authors of "OpenAI's nonprofit governance structure" for outstanding contributions to creative fiction.

PauseAI US's Executive Director blocked Liv, and has since then been repeatedly quote-tweeting her to criticise her. PauseAI folk, if these aren't your values or behaviour you think is reflective of what Pause AI is or should be, now is the time to say.

"Losing oversight and control of advanced AI systems, particularly Artificial General Intelligence (AGI), would be catastrophic. It must be avoided at all costs." Peter Kyle, UK Technology Secretary

Up until 2025, what i would hear was "we can't afford to invest in safety because China won't invest in safety". Now that it's increasingly clear that the appetite for safety and safety-focused international governance is there for China, I fully expect to hear the opposite: if Chinese voices are calling for it, it can't be in 'our' interests to do it. My opinion: powerful AI interests with deep pockets don't want to be subject to governance, oversight, or safety requirements. Make them be.

This has been coming. Now it's arriving. I need to dig into the details, but it could be very noteworthy. I discussed my concerns about the pace of change in AI-assisted/enhanced cyberattack capabilities in a short paper that'll hopefully be out next week. Seems I might already be out of date.

Let me say one thing that actually matters: if we believe their staff then what we're seeing, and that we're seeing eval results for, is not OpenAI's most capable model. They have more capable in-house. Without transparency requirements, this asymmetry will only grow, and will make meaningful awareness and governance increasingly ineffective + eventually impossible. The community needs to keep pushing for transparency requirements and testing of models in development and deployed internally.

Flew from Beijing Daixing for the first time today. It's a genuinely beautiful and thoughtfully designed building, and the effect on mood, and pleasantness of the airport experience, is noticeable. Photo is of the garden next to my departure gate. I sometimes feel we're being robbed in the ugliness of our infrastructure. The British managed it in the 19th Century (see the glory that is St Pancras) and the Chinese are managing it in the 21st Century, so why can't this be more common?

my kind of transhumanist. (quietly admiring David Pearce fighting the good fight on twitter)

One of the weird things about the world today is that the idea of 'AGI' is now regularly being talked about in e.g. policy contexts. But seems v clear that most policymakers' notion of AGI & its implications is vastly underpowered compared to that of the ppl trying to build AGI.