Emergent Capabilities

Amodei: smooth scaling hides sudden emergence — the three-digit example

Dario Amodei

“Although loss curves improve smoothly with scale, individual capabilities can flip from absent to present almost overnight. Amodei illustrates this with arithmetic on three-digit numbers: a model at one scale simply cannot do it, then at the next scale it can — with no warning from the aggregate metrics.”

Sutskever: reliability and controllability as desirable emergent properties

Ilya Sutskever

“Not all emergence is dangerous — Sutskever argues that reliability and controllability are themselves emergent properties we want to appear at scale, and that understanding the conditions under which beneficial properties emerge is a core research challenge.”

Amodei: today's emergent capabilities as a laboratory for future safety

Dario Amodei

“The emergent capabilities we observe in today's models — surprising, hard to predict, hard to explain — are a preview of what much more powerful systems will exhibit. Amodei frames the current moment as a unique opportunity to study emergence while the stakes are still low enough to experiment safely.”

Fully autonomous robots are much closer than you think – Sergey Levine

Sergey Levine

“That's actually where the emergent capabilities come from. Because of this, in principle, if we have a sufficient diversity of behaviors, the model should figure out that those behaviors can be composed in new ways as the situation calls for it. We've actually”

Mutually Assured AI Malfunction [Dan Hendrycks]

Dan Hendrycks

“…ke 2021 or something like that, I use the phrase emergence, emergent capabilities. I believe I may have been the the first in in the literature to use it. So, I feel it was later used by Jason Wei's sort of thing or Jacob Steinhard did in a blog post, my advis”

Fully autonomous robots are much closer than you think – Sergey Levine

Sergey Levine

“…progress more generally than that? There's a subtlety here. Emergent capabilities don't just come from the fact that internet data has a lot of stuff in it. They also come from the fact that generalization, once it reaches a certain level, becomes compositiona”

Darren McKee on Uncontrollable Superintelligence

Darren McKee

“…maybe so that we should be very careful there's a bunch of emergent capabilities in AI systems that could be problematic if we're trying to align these systems so so we're talking about power seeking deception man manipulation of humans using humans to achiev”

Roman Yampolskiy: Dangers of Superintelligent AI | Lex Fridman Podcast #431

Roman Yampolskiy

“…t, to you, the difference there that there is some level of emergent intelligence that happens in our current approaches? So stuff that we don't hardcode in. - Absolutely. That's what makes it so successful. When we had to painstakingly hardcode in everything,”

Large language models, OpenAI, and striving to make the future go well | Richard Ngo

Richard Ngo

“…that we just haven't figured out yet, because they're like emergent properties of what the model has learned during training. So probably the best example of this is that people figured out that if you prompt the model to think through step by step in its rea”

Oriol Vinyals: Deep Learning and Artificial General Intelligence | Lex Fridman Podcast #306

Oriol Vinyals

“…lly has the same abilities you can teach it for instance an emergent property was that you can take pictures of numbers and then do do arithmetic with the numbers just by teaching it oh that's i mean when when i show you three plus six you know i want you to o”

#31 - Prof Dafoe on defusing the political & economic risks posed by existing AI capabilities

80,000 Hours

“…pursuing their interests as they understand it given their capabilities and try and understand how that all plays out and how we can address and mitigate the worst potential risks from that and seize the opportunities seize the opportunities seize the opportu”