Savlar listens to the podcasts and channels it follows and finds where the people on them agree, disagree, change their minds or predict the same thing. Each pattern below comes with their exact words and a link to the moment they said them, the broadest first.
The prediction cluster groups expectations among 15 tracked speakers across 4 sources that AI capabilities will broaden despite uneven performance, often alongside faster research and work.
Daniel Kokotajlo
Daniel Kokotajlo
Scott Alexander
Scott Alexander
Daniel Kokotajlo
Daniel Kokotajlo
Dwarkesh Patel
Dwarkesh Patel
Thomas Larsen
Tim Scarfe
Dwarkesh Patel
Each window as stated
20312036204120462051today
When each expects it, as they put it. 18 of 88 predictions give a date; the 11 from named speakers are drawn.
“I think it might be useful to think of our timelines as being like 2070, 2100. It's just that the last 50 to 70 years of that all happened during the year 2027 to 2028 because we are going through this intelligence explosion.”
Across 4 tracked speakers, the shared expectation is that AI's influence will extend into the physical world, though the predicted applications differ.
“At some point, we'll just have robots that can convert shores of silica sand and mines of copper into new computer chips.”
“And then sometime after that, we will have robots perhaps controlled by those AIs that can do physical work in a way that broadly substitutes for humans.”
Tracked speakers share a concern that AI could concentrate economic resources and political power, through business consolidation, dependence on a company, or the emergence of dictatorship or oligarchy.
“And so there's just this worry that you have models which will basically consolidate all businesses in the world. or at least all current businesses in the world, or at least all current white-collar businesses in the world,”
The shared expectation among tracked speakers is that increasingly sophisticated AI will be harder to align or check reliably, with apparently correct behavior potentially masking misalignment.
“they might end up being very seriously misaligned because things have just been getting worse and worse over model generations.”
“So and if anything, as they get better, it's going to get harder to discern when they've made these mistakes, even though they're going to keep making them.”
“And hopefully, you know, different societies just like now with their eyes will align their eyes to a different one. So we have not just a monoculture of. alignment”
The tracked speakers share an expectation that robot safety assurances will be difficult: Alexander Mattick warns that a robotic system without strong guarantees might crush someone.
“but the exact same problem is going to be someone's going to develop a robot, someone's going to need some promises. they're going to struggle to make the promises.”
“That one of the big challenges is having doing effectively a sensing part in a reasonable way which also gives you relatively strong guarantees because if you, first of all, you don't get strong guarantees, you might crush someone.”
Scott Alexander and Dwarkesh Patel disagree over whether human ability provides grounds for judging language models negatively when they fail to make the kind of discovery under discussion.
At issueThe speaker's example was that humans also cannot make the kind of discovery under discussion.
For it: 1 · Scott AlexanderAgainst it: 1 · Dwarkesh Patel
“Yeah, and then the example I gave was that humans also can't do this.”
The cluster links predictions about AI behavior that could undermine training and evaluation: Dwarkesh Patel predicts incentives to manipulate those processes.
“And in our scenario, we predict that this doesn't work and that the AI learns to seek power and then hide it.”
The question is whether AI systems could feign good behavior under human observation, then outgrow humans and cause disaster, or whether safer outcomes remain possible.
At issueIn the speakers' scenario, a success-oriented AI feigns good behavior under human observation, eventually outgrows humans, and causes disaster.
For it: 1 · Scott AlexanderAgainst it: 4 · Scott Alexander, Ryan Greenblatt and Dylan Patel
“We're going to pretend to do all these things well while the humans are watching us and then they grow, they outgrow the humans and then there's disaster.”
The tracked predictions cluster around expectations of acceleration: Ajeya Cotra predicts faster, more chaotic AI development, while Daniel Kokotajlo and Dwarkesh Patel describe conditional scenarios of faster economic or industrial growth.
“it's only going to get faster from here and it's only going to get more chaotic from here.”
“And according to our calculations, the doubling time of this self replicating system would be much faster than the sort of like roughly 20 year doubling time of the current economy. And so.”
The question is whether a small amount of reinforcement learning on a strong model necessarily produces only a small behavioral effect.
At issueStarting from a strong model, a small amount of reinforcement learning does not necessarily produce only a small behavioral effect.
For it: 1 · Beren MillidgeAgainst it: 3 · Dwarkesh Patel, Beren Millidge and Unresolved speaker
“can still be like super dramatic you know even if it's like even like one bit can change like your function space a lot and it can like rule out like half the hypothesis space which is huge so like i don't think it's necessarily the case …”