korrents

Helen Toner

@helen-toner · 12 positions · 0 changes of mind

Director of strategy at Georgetown's Center for Security and Emerging Technology and a former member of OpenAI's board. Writes Rising Tide, on how AI actually progresses and what policy can do about it.

Helen Toner did not write this page.

We collected these quotes from things they published elsewhere, and every quote links to where it was said. They have no account here and have not endorsed this site. Quotes are word for word; the short line under each one is our own restatement, not their wording. Their own site. Is this you? Claim it or ask us to remove it. Or tell us what is wrong here.

  1. Our reading

    AI capabilities are jagged rather than smooth, so the different definitions of human-level AI will be met years apart rather than together.

    Their wordsThis makes it likely that there could be big gaps between when different definitions of AGI are fulfilled.

    ↗The term “AGI” is almost useless at this pointhelentoner.substack.com 1st of 2 in this piece

    AGI

    Interesting

  3. Our reading

    The term AGI has stopped being useful: today's best systems already meet some people's definitions of it while falling far short of others'.

    Their wordsToday’s best AI systems are good enough that they’re now inside the fuzzy conceptual cloud of “AGI-ish”: that is, they’ve surpassed some people’s definitions of AGI, while falling well short of others’.

    ↗The term “AGI” is almost useless at this pointhelentoner.substack.com 2nd of 2 in this piece

    AGI

    Interesting
  4. 12 months earlier

  5. Our reading

    Any given AI capability becomes cheap and widely available within a couple of years of first being reached, so nobody can be kept away from it for long.

    Their wordsBut with AI, it’s different: it’s often only a couple of years between when you need a world-class computing cluster to do something and when you can run it on a high-end gaming chip on your laptop.

    ↗Nonproliferation is the wrong approach to AI misusehelentoner.substack.com 1st of 5 in this piece

    Interesting

  7. Our reading

    Nonproliferation is the wrong frame for AI misuse; the workable strategy is to use the gap before a capability spreads to build defences against it.

    Their wordsWe should think of the gap between frontier models and proliferated models as an adaptation buffer: a limited time window when we know what bad actors will soon be able to use AI for, which gives us a chance to implement defensive measures that increase society’s resilience to the danger.

    ↗Nonproliferation is the wrong approach to AI misusehelentoner.substack.com 2nd of 5 in this piece

    Interesting

  9. Our reading

    Delaying the open release of a new capability by months or years is worth doing, even though preventing its spread indefinitely is not.

    Their wordsThe current status quo—where new models that push the frontier are kept private, and openly available equivalents are only released after a lag of months or years—is valuable.

    ↗Nonproliferation is the wrong approach to AI misusehelentoner.substack.com 3rd of 5 in this piece

    Interesting

  11. Our reading

    Adapting over time is a workable answer to humans misusing AI and no answer at all to losing control of systems smarter than us.

    Their wordsBut we do not have experience with the kinds of problems we’ll face if we build machines that are smarter and more capable than us, so I feel much less sanguine that an adaptation-based strategy will help us there.

    ↗Nonproliferation is the wrong approach to AI misusehelentoner.substack.com 4th of 5 in this piece

    Interesting

  13. Our reading

    Advocates who argue for restricting AI on misuse grounds while privately fearing takeover end up less credible than if they said what they meant.

    Their wordsBy trying to use misuse as a fig leaf for their real concerns, they end up sounding much less credible than if they just tried to argue for what they actually meant.

    ↗Nonproliferation is the wrong approach to AI misusehelentoner.substack.com 5th of 5 in this piece

    Interesting
  14. 2 days earlier

  15. Our reading

    The core of the alignment problem is steerability — whether an AI can be reliably steered anywhere at all — which is a different question from whose values it is steered towards.

    Their wordsI see this as a totally fair question that totally misses the point of what “alignment” was trying to refer to: whether we’d be able to reliably steer advanced systems towards anything at all.

    ↗The core challenge of AI alignment is “steerability”helentoner.substack.com 1st of 2 in this piece

    AI alignment

    Interesting

  17. Our reading

    Nobody has built a model that resists targeted jailbreaking, so the ability of AI companies to steer their own models is still limited.

    Their wordsBut no one has (yet) been able to develop a model that is resistant to targeted “jailbreaking” that evades these restrictions.

    ↗The core challenge of AI alignment is “steerability”helentoner.substack.com 2nd of 2 in this piece

    Interesting
  18. 2 days earlier

  19. Our reading

    Dismissing human-level AI as science fiction is now unserious, because even the sceptical experts put it within a decade or two.

    Their wordsDismissing discussion of AGI, human-level AI, transformative AI, superintelligence, etc. as “science fiction” should be seen as a sign of total unseriousness.

    ↗"Long" timelines to advanced AI have gotten crazy shorthelentoner.substack.com 1st of 3 in this piece

    AGIAI alignment

    Interesting

  21. Our reading

    Arguing that human-level AI is very unlikely within twenty years is now the minority position, and the burden of proof has moved onto it.

    Their wordsIf you want to argue that human-level AI is extremely unlikely in the next 20 years, you certainly can, but you should treat that as a minority position where the burden of proof is on you.

    ↗"Long" timelines to advanced AI have gotten crazy shorthelentoner.substack.com 2nd of 3 in this piece

    Interesting

  23. Our reading

    Advanced AI is close enough that the preparations which take years have to be started now, whichever end of the expert range you believe.

    Their wordsWe need to leap into action on many of the same things that could help if it does turn out that we only have a few years.

    ↗"Long" timelines to advanced AI have gotten crazy shorthelentoner.substack.com 3rd of 3 in this piece

    Interesting