id
stringlengths
17
17
body
stringlengths
0
19.5k
posted_at
stringlengths
24
24
karma
int64
-6
185
parent_comment_id
stringlengths
17
17
post_id
stringlengths
17
17
post_title
stringlengths
2
127
post_slug
stringlengths
2
61
post_url
stringlengths
20
146
post_author
stringclasses
214 values
post_posted_at
stringlengths
24
24
oHBLWWFckri4P3B37
Consider tasks that quite good software engineers (maybe top 40% at Jane Street) typically do in 8 hours without substantial prior context on that exact task. (As in, 8 hour median completion time.) Now, we'll aim to sample these tasks such that the distribution and characteristics of these tasks are close to the distr...
2025-02-13T23:36:50.331Z
16
bA7miafttNQkLckqw
gsj3TWdcBxwkm9eNt
≤10-year Timelines Remain Unlikely Despite DeepSeek and o3
10-year-timelines-remain-unlikely-despite-deepseek-and-o3
https://www.lesswrong.com/posts/gsj3TWdcBxwkm9eNt/10-year-timelines-remain-unlikely-despite-deepseek-and-o3
Rafael Harth
2025-02-13T19:21:35.392Z
pwqSm7fqtDSxyuQRt
I think if you look at "horizon length"---at what task duration (in terms of human completion time) do the AIs get the task right 50% of the time---the trends will indicate doubling times of maybe 4 months (though 6 months is plausible). Let's say 6 months more conservatively. I think AIs are at like 30 minutes on math...
2025-02-13T23:46:13.830Z
12
M5irFzmnddJjHFmWy
gsj3TWdcBxwkm9eNt
≤10-year Timelines Remain Unlikely Despite DeepSeek and o3
10-year-timelines-remain-unlikely-despite-deepseek-and-o3
https://www.lesswrong.com/posts/gsj3TWdcBxwkm9eNt/10-year-timelines-remain-unlikely-despite-deepseek-and-o3
Rafael Harth
2025-02-13T19:21:35.392Z
vtYtRtCcsoEiivcAH
Importantly, this is an example of developing a specific application (surgical robot) rather than advancing the overall field (robots in general). It's unclear whether the analogy to an individual application or an overall field is more appropriate for AI safety.
2025-02-13T23:47:16.468Z
2
nzHmPi8WtFnpRiCbd
Hgj84BSitfSQnfwW6
So You Want To Make Marginal Progress...
so-you-want-to-make-marginal-progress
https://www.lesswrong.com/posts/Hgj84BSitfSQnfwW6/so-you-want-to-make-marginal-progress
johnswentworth
2025-02-07T23:22:19.825Z
knwahJiREnLT7dNfo
No, sorry I was mostly focused on "such that if you didn't see them within 3 or 5 years, you'd majorly update about time to the type of AGI that might kill everyone". I didn't actually pick up on "most impressive" and actually tried to focus on something that occurs substantially before things get crazy. Most impressi...
2025-02-13T23:50:16.206Z
9
fycZ8fjqHzXbyoCZb
gsj3TWdcBxwkm9eNt
≤10-year Timelines Remain Unlikely Despite DeepSeek and o3
10-year-timelines-remain-unlikely-despite-deepseek-and-o3
https://www.lesswrong.com/posts/gsj3TWdcBxwkm9eNt/10-year-timelines-remain-unlikely-despite-deepseek-and-o3
Rafael Harth
2025-02-13T19:21:35.392Z
eABhutT7yKt6NfDLH
I think you can add mirror enzymes which can break down mirror carbs. Minimally we are aware of enzymes which break down mirror glucose.
2025-02-13T23:57:30.569Z
3
mar9ZByKGrPpSShox
KFJ2LFogYqzfGB3uX
How AI Takeover Might Happen in 2 Years
how-ai-takeover-might-happen-in-2-years
https://x.com/joshua_clymer/status/1887905375082656117
joshc
2025-02-07T17:10:10.530Z
aawHBx4eTPXJGtEud
I mean, I don't think AI R&D is a particularly hard field persay, but I do think it involves lots of tricky stuff and isn't much easier than automating some other plausibly-important-to-takeover field (e.g., robotics). (I could imagine that the AIs have a harder time automating philosophy even if they were trying to wo...
2025-02-14T00:01:44.613Z
9
Mcwmrg9PBvNuXAqRQ
gsj3TWdcBxwkm9eNt
≤10-year Timelines Remain Unlikely Despite DeepSeek and o3
10-year-timelines-remain-unlikely-despite-deepseek-and-o3
https://www.lesswrong.com/posts/gsj3TWdcBxwkm9eNt/10-year-timelines-remain-unlikely-despite-deepseek-and-o3
Rafael Harth
2025-02-13T19:21:35.392Z
QtxDyySpgxajDEYdp
The question of context might be important, see [here](https://www.lesswrong.com/posts/gsj3TWdcBxwkm9eNt/10-year-timelines-remain-unlikely-despite-deepseek-and-o3?commentId=oHBLWWFckri4P3B37). I wouldn't find 15 minutes that surprising for ~50% success rate, but I've seen numbers more like 1.5 hours. I thought this was...
2025-02-14T00:58:50.339Z
7
MrGPFJ7f7QekZ2Rvh
gsj3TWdcBxwkm9eNt
≤10-year Timelines Remain Unlikely Despite DeepSeek and o3
10-year-timelines-remain-unlikely-despite-deepseek-and-o3
https://www.lesswrong.com/posts/gsj3TWdcBxwkm9eNt/10-year-timelines-remain-unlikely-despite-deepseek-and-o3
Rafael Harth
2025-02-13T19:21:35.392Z
Ke45eerx8kTiK9QF3
(I don't expect o3-mini is a much better agent than 3.5 sonnet new out of the box, but probably a hybrid scaffold with o3 + 3.5 sonnet will be substantially better than 3.5 sonnet. Just o3 might also be very good. Putting aside cost, I think o1 is usually better than o3-mini on open ended programing agency tasks I thin...
2025-02-14T01:47:36.372Z
6
wvDiPKhLSarLcW5cT
gsj3TWdcBxwkm9eNt
≤10-year Timelines Remain Unlikely Despite DeepSeek and o3
10-year-timelines-remain-unlikely-despite-deepseek-and-o3
https://www.lesswrong.com/posts/gsj3TWdcBxwkm9eNt/10-year-timelines-remain-unlikely-despite-deepseek-and-o3
Rafael Harth
2025-02-13T19:21:35.392Z
fQkr7AaepGNiFjbRv
> Some things are true simply because they are true and in general there's no reason to expect a simpler explanation. You could believe: Some things are true simply because they are true, but only when being true isn't very surprising. (For instance, it isn't very surprising that there are some cellular automata that...
2025-02-18T00:10:03.622Z
9
qaLYLmswHv5H8GkpJ
Xt9r4SNNuYxW83tmo
A computational no-coincidence principle
a-computational-no-coincidence-principle
https://www.alignment.org/blog/a-computational-no-coincidence-principle/
Eric Neyman
2025-02-14T21:39:39.277Z
z89CmZXfDT8j5Rtci
Note that the capability milestone forecasted in the linked short form is substantially weaker than the notion of transformative AI in the 2020 model. (It was defined as AI with an effect at least as large as the industrial revolution.) I don't expect this adds many years, for me it adds like ~2 years to my median. (...
2025-02-20T00:58:48.782Z
10
7bZGiKqmLbMgumH5A
TTFsKxQThrqgWeXYJ
How might we safely pass the buck to AI?
how-might-we-safely-pass-the-buck-to-ai
https://www.lesswrong.com/posts/TTFsKxQThrqgWeXYJ/how-might-we-safely-pass-the-buck-to-ai
joshc
2025-02-19T17:48:32.249Z
j3KEdDjMLGSJCTtp7
I think there is an important component of trustworthiness that you don't emphasize enough. It isn't sufficient to just rule out alignment faking, we need the AI to actually try hard to faithfully pursue our interests including on long, confusing, open-ended, and hard to check tasks. You discuss establishing this with ...
2025-02-20T03:21:37.535Z
24
null
TTFsKxQThrqgWeXYJ
How might we safely pass the buck to AI?
how-might-we-safely-pass-the-buck-to-ai
https://www.lesswrong.com/posts/TTFsKxQThrqgWeXYJ/how-might-we-safely-pass-the-buck-to-ai
joshc
2025-02-19T17:48:32.249Z
ZwPBae2EdimyXgiLz
But it isn't a capabilities condition? Maybe I would be happier if you renamed this section.
2025-02-20T04:32:16.098Z
4
P94ttWmdzFWPE4ycg
TTFsKxQThrqgWeXYJ
How might we safely pass the buck to AI?
how-might-we-safely-pass-the-buck-to-ai
https://www.lesswrong.com/posts/TTFsKxQThrqgWeXYJ/how-might-we-safely-pass-the-buck-to-ai
joshc
2025-02-19T17:48:32.249Z
SzPwzMz96qwCAyN3x
To be clear, I think there are important additional considerations related to the fact that we don't just care about capabilities that aren't covered in that section, though that section is not that far from what I would say if you renamed it to "behavioral tests", including both capabilities and alignment (that is, al...
2025-02-20T07:46:01.619Z
2
ZwPBae2EdimyXgiLz
TTFsKxQThrqgWeXYJ
How might we safely pass the buck to AI?
how-might-we-safely-pass-the-buck-to-ai
https://www.lesswrong.com/posts/TTFsKxQThrqgWeXYJ/how-might-we-safely-pass-the-buck-to-ai
joshc
2025-02-19T17:48:32.249Z
2o8475t3hCHoNTbZa
Something important is that "significantly accelerate alignment research" isn't the same as "making AIs that we're happy to fully defer to". This post is talking about conditions needed for deference and how we might achieve them. (Some) acceleration doesn't require being fully competitive with humans while deference ...
2025-02-25T01:54:08.420Z
9
NP5Puxne6Njp5mbRm
5gmALpCetyjkSPEDr
Training AI to do alignment research we don’t already know how to do
training-ai-to-do-alignment-research-we-don-t-already-know
https://www.lesswrong.com/posts/5gmALpCetyjkSPEDr/training-ai-to-do-alignment-research-we-don-t-already-know
joshc
2025-02-24T19:19:43.067Z
WckskPDPhY3kojg9t
FWIW, I don't think "data-efficient long-horizon RL" (which is sample efficient in a online training sense) implies you can make faithful simulations. > but you do get a machine that pursues the metric that you fine-tuned it to pursue, even out of distribution (with a relatively small amount of data). I think if the ...
2025-02-25T22:37:13.659Z
4
aLBiFt2t8i9Y2wfid
5gmALpCetyjkSPEDr
Training AI to do alignment research we don’t already know how to do
training-ai-to-do-alignment-research-we-don-t-already-know
https://www.lesswrong.com/posts/5gmALpCetyjkSPEDr/training-ai-to-do-alignment-research-we-don-t-already-know
joshc
2025-02-24T19:19:43.067Z
2GdrZuNh9fbdF5FWa
A typical crux is that I think we can increase our chances of "real alignment" using prosaic and relatively unenlightened ML reasearch without any deep understanding. I both think: 1. We can significantly accelerate prosaic ML safety research (e.g., of the sort people are doing today) using AIs that are importantly l...
2025-02-25T22:40:24.188Z
3
WNsrDf3JW6XjDrcDK
5gmALpCetyjkSPEDr
Training AI to do alignment research we don’t already know how to do
training-ai-to-do-alignment-research-we-don-t-already-know
https://www.lesswrong.com/posts/5gmALpCetyjkSPEDr/training-ai-to-do-alignment-research-we-don-t-already-know
joshc
2025-02-24T19:19:43.067Z
mouqe5fkSFhrwfPHB
I don't think "what is the necessary work for solving alignment" is a frame I really buy. My perspective on alignment is more like: - Avoiding egregious misalignment (where AIs intentionally act in ways that make our tests highly misleading or do [pretty obviously unintended/dangerous actions](https://www.lesswrong.co...
2025-02-26T01:20:42.410Z
3
Mf8ASbnBM66pKtGEt
5gmALpCetyjkSPEDr
Training AI to do alignment research we don’t already know how to do
training-ai-to-do-alignment-research-we-don-t-already-know
https://www.lesswrong.com/posts/5gmALpCetyjkSPEDr/training-ai-to-do-alignment-research-we-don-t-already-know
joshc
2025-02-24T19:19:43.067Z
gomb2WAdKdnDpGZeS
Oh, yeah I meant "perform well according to your metrics" not "behave well" (edited)
2025-02-26T02:28:22.065Z
2
2Bf42LQ8SBQAaasvd
5gmALpCetyjkSPEDr
Training AI to do alignment research we don’t already know how to do
training-ai-to-do-alignment-research-we-don-t-already-know
https://www.lesswrong.com/posts/5gmALpCetyjkSPEDr/training-ai-to-do-alignment-research-we-don-t-already-know
joshc
2025-02-24T19:19:43.067Z
4ipCySAZ3WKNycadt
> On some axes, but won't there to be axes where AIs are more difficult than humans also? Sycophancy&slop being the most salient. Misalignment issues being another. Yes, I just meant on net. (Relative to the current ML community and given a similar fraction of resources to spend on AI compute.)
2025-02-26T04:13:08.437Z
2
TXQK4ECryYmKiZxwR
5gmALpCetyjkSPEDr
Training AI to do alignment research we don’t already know how to do
training-ai-to-do-alignment-research-we-don-t-already-know
https://www.lesswrong.com/posts/5gmALpCetyjkSPEDr/training-ai-to-do-alignment-research-we-don-t-already-know
joshc
2025-02-24T19:19:43.067Z
vaParfzSGXoAa3jiH
I certainly agree it isn't clear, just my current best guess.
2025-02-26T05:02:15.662Z
4
N9EqDigK48D9tzKku
5gmALpCetyjkSPEDr
Training AI to do alignment research we don’t already know how to do
training-ai-to-do-alignment-research-we-don-t-already-know
https://www.lesswrong.com/posts/5gmALpCetyjkSPEDr/training-ai-to-do-alignment-research-we-don-t-already-know
joshc
2025-02-24T19:19:43.067Z
jChY95BeDeptDpnZK
I think something like this is a live concern, though I'm skeptical that control is net negative for this reason. My baseline guess is that trying to detect AIs doing problematic actions makes it more likely that we get evidence for misalignment that triggers a useful response from various groups. I think it would be ...
2025-03-11T19:18:43.349Z
56
null
rZcyemEpBHgb2hqLP
AI Control May Increase Existential Risk
ai-control-may-increase-existential-risk
https://www.lesswrong.com/posts/rZcyemEpBHgb2hqLP/ai-control-may-increase-existential-risk
Jan_Kulveit
2025-03-11T14:30:05.972Z
ekegtfZ7nxvDonWrp
> I do understand this line of reasoning, but yes, my intuition differs. For some sort of a weird case study, consider Sydney. [...] My guess is that none of that would have happened with properly implemented control measures. Sure, or with properly implemented ~anything related to controlling the AIs behavior. I don...
2025-03-12T16:16:01.280Z
6
Sx6xczrvwirTnXHRp
rZcyemEpBHgb2hqLP
AI Control May Increase Existential Risk
ai-control-may-increase-existential-risk
https://www.lesswrong.com/posts/rZcyemEpBHgb2hqLP/ai-control-may-increase-existential-risk
Jan_Kulveit
2025-03-11T14:30:05.972Z
irkfgCzSMu5L9P3Ld
Related question: are you in favor of making AGI open weights? By AGI, I mean AIs which effectively operate autonomously and can acquire automously acquire money/power. This includes AIs capable enough to automate whole fields of R&D (but not much more capable than this). I think the case for this being useful on your ...
2025-03-12T16:38:48.742Z
6
null
rZcyemEpBHgb2hqLP
AI Control May Increase Existential Risk
ai-control-may-increase-existential-risk
https://www.lesswrong.com/posts/rZcyemEpBHgb2hqLP/ai-control-may-increase-existential-risk
Jan_Kulveit
2025-03-11T14:30:05.972Z
bEnBYNQK5HbCugCvk
You could have the view that open weights AGI is too costly on takeover risk and escape is bad, but we'll hopefully have some pre-AGI AIs which do strange misaligned behaviors that don't really get them much/any influence/power. If this is the view, then it really feels to me like preventing escape/rogue internal deplo...
2025-03-12T17:06:01.883Z
6
irkfgCzSMu5L9P3Ld
rZcyemEpBHgb2hqLP
AI Control May Increase Existential Risk
ai-control-may-increase-existential-risk
https://www.lesswrong.com/posts/rZcyemEpBHgb2hqLP/ai-control-may-increase-existential-risk
Jan_Kulveit
2025-03-11T14:30:05.972Z
bqdfaTaPz22ZnsHAt
My guess is that the parts of the core leadership of Anthropic which are thinking actively about misalignment risks (in particular, Dario and Jared) think that misalignment risk is like ~5x smaller than I think it is while also thinking that risks from totalitarian regimes are like 2x worse than I think they are. I thi...
2025-03-14T23:27:04.929Z
37
null
7uTPrqZ3xQntwQgYz
Anthropic, and taking "technical philosophy" more seriously
anthropic-and-taking-technical-philosophy-more-seriously
https://www.lesswrong.com/posts/7uTPrqZ3xQntwQgYz/anthropic-and-taking-technical-philosophy-more-seriously
Raemon
2025-03-13T01:48:54.184Z
vyDvoWaHxKfBcffEy
> Dario Amodei says AI will be writing 90% of the code in 6 months and almost all the code in 12 months. I think it's somewhat unclear how big of a deal this is. In particular, situations where AIs write 90% of lines of code, but are very far (in time, effective compute, and qualitative capabilities) from being able t...
2025-03-15T00:56:22.418Z
9
null
XFGTJz9vGwjJADeFB
AI #107: The Misplaced Hype Machine
ai-107-the-misplaced-hype-machine
https://www.lesswrong.com/posts/XFGTJz9vGwjJADeFB/ai-107-the-misplaced-hype-machine
Zvi
2025-03-13T14:40:05.318Z
Yc5mCkpBS8aLv53f7
I don't expect 90% of code in 6 months and more confidently don't expect "almost all" in 12 months for a reasonable interpretation of almost all. However, I think this prediction is also weaker than it might seem, see [my comment here](https://www.lesswrong.com/posts/XFGTJz9vGwjJADeFB/ai-107-the-misplaced-hype-machine?...
2025-03-15T01:02:34.950Z
5
GXtpNwunm6XGWTfav
XFGTJz9vGwjJADeFB
AI #107: The Misplaced Hype Machine
ai-107-the-misplaced-hype-machine
https://www.lesswrong.com/posts/XFGTJz9vGwjJADeFB/ai-107-the-misplaced-hype-machine
Zvi
2025-03-13T14:40:05.318Z
TeofHxezeNfKWyMxh
> IMO actively torch the "long pause" worlds Not sure how interesting this is to discuss, but I don't think I agree with this. Stuff they're doing does seem harmful to worlds where you need a long pause, but feels like at the very least Anthropic is a small fraction of the torching right? Like if you think Anthropic i...
2025-03-15T04:10:15.688Z
5
rxtpytGBquCwEAMp7
7uTPrqZ3xQntwQgYz
Anthropic, and taking "technical philosophy" more seriously
anthropic-and-taking-technical-philosophy-more-seriously
https://www.lesswrong.com/posts/7uTPrqZ3xQntwQgYz/anthropic-and-taking-technical-philosophy-more-seriously
Raemon
2025-03-13T01:48:54.184Z
frp73opG8oTogtxBd
Yep, just the obvious. (I'd say "much less bought in" than "isn't bought in", but whatever.) I don't really have dots I'm trying to connect here, but this feels more central to me than what you discuss. Like, I think "alignment might be really, really hard" (which you focus on) is less of the crux than "is misalignmen...
2025-03-15T04:17:21.912Z
4
DXdq9ab3xScHsFuFy
7uTPrqZ3xQntwQgYz
Anthropic, and taking "technical philosophy" more seriously
anthropic-and-taking-technical-philosophy-more-seriously
https://www.lesswrong.com/posts/7uTPrqZ3xQntwQgYz/anthropic-and-taking-technical-philosophy-more-seriously
Raemon
2025-03-13T01:48:54.184Z
gPqiyqfFTssBAjaKH
> The "extreme philosophical competence" hypothesis is that you need such competence to achieve "seriously aligned" in this sense. It sounds like you disagree, but I don't know why since your reasoning just sidesteps the problem. Yes, my reasoning is definitely part, but not all of the argument. Like the thing I said ...
2025-03-15T05:09:22.858Z
14
2jBvsucWWM7MTvsuM
7uTPrqZ3xQntwQgYz
Anthropic, and taking "technical philosophy" more seriously
anthropic-and-taking-technical-philosophy-more-seriously
https://www.lesswrong.com/posts/7uTPrqZ3xQntwQgYz/anthropic-and-taking-technical-philosophy-more-seriously
Raemon
2025-03-13T01:48:54.184Z
uXkW78Kykz4H4ansQ
> Orcas have about 43 billion cortical neurons - humans have about 21 billion. The orca cortex has 6 times the area of the human cortex, though the neuron density is about 3 times lower. > > [...] > > My uncertain guess is that, within mammalian brains, scaling matters a lot more for individual intelligence, This post...
2025-03-16T20:25:18.694Z
14
null
dzLwCBvwC4hWytnus
Considerations on orca intelligence
considerations-on-orca-intelligence
https://www.lesswrong.com/posts/dzLwCBvwC4hWytnus/considerations-on-orca-intelligence
Towards_Keeperhood
2024-12-29T14:35:16.445Z
ugLqGyNJCiJFJZWaY
> This makes sense as a crux for the claim "we need philosophical competence to align unboundedly intelligent superintelligences." But, it doesn't make sense for the claim "we need philosophical competence to align general, openended intelligence." I was thinking of a slightly broader claim: "we need extreme philosoph...
2025-03-16T20:47:49.519Z
2
rE2RnTiv4i9mqwECe
7uTPrqZ3xQntwQgYz
Anthropic, and taking "technical philosophy" more seriously
anthropic-and-taking-technical-philosophy-more-seriously
https://www.lesswrong.com/posts/7uTPrqZ3xQntwQgYz/anthropic-and-taking-technical-philosophy-more-seriously
Raemon
2025-03-13T01:48:54.184Z
g7YyfoiCovX4P3SWL
Yes, I was intending my comment to refer to just code at Anthropic. (Otherwise I would talk much more about serious integration lags and lack of compute.)
2025-03-17T15:39:15.034Z
5
CB6rcgYyhn8yCfYok
XFGTJz9vGwjJADeFB
AI #107: The Misplaced Hype Machine
ai-107-the-misplaced-hype-machine
https://www.lesswrong.com/posts/XFGTJz9vGwjJADeFB/ai-107-the-misplaced-hype-machine
Zvi
2025-03-13T14:40:05.318Z
RcKjDmZbtQcamqpuP
Some quick (and relatively minor) notes: - I expect that full stack intelligence explosion could look more like "make the whole economy bigger using a bunch of AI labor" rather than specifically automating the chip production process. (That said, in practice I expect explicit focused automation of chip production to b...
2025-03-17T18:24:14.708Z
28
null
PzbEpSGvwH3NnegDB
Three Types of Intelligence Explosion
three-types-of-intelligence-explosion
https://www.forethought.org/research/three-types-of-intelligence-explosion
rosehadshar
2025-03-17T14:47:46.696Z
K9qoqBNBnFB7r7oMW
> Number cortical neurons != brain size. Orcas have ~2x the number of cortical neurons, but much larger brains. Assuming brain weight is proportional to volume, with human brains being typically 1.2-1.4kg, and orca brains being typically 5.4-6.8kg, orca brains are actually like 6.1/1.3=4.7 times larger than human brain...
2025-03-17T18:40:35.922Z
4
FaBd8sM8T9ax3cJsM
dzLwCBvwC4hWytnus
Considerations on orca intelligence
considerations-on-orca-intelligence
https://www.lesswrong.com/posts/dzLwCBvwC4hWytnus/considerations-on-orca-intelligence
Towards_Keeperhood
2024-12-29T14:35:16.445Z
3Dhs5jZpBimxS9gz7
It looks like the images aren't showing up on LW.
2025-03-20T02:29:34.750Z
2
null
ZhDtfuBnxhC6X8ycG
Improved visualizations of METR Time Horizons paper.
improved-visualizations-of-metr-time-horizons-paper
https://www.lesswrong.com/posts/ZhDtfuBnxhC6X8ycG/improved-visualizations-of-metr-time-horizons-paper
LDJ
2025-03-19T23:36:52.771Z
C3H7LgZcFk5eu2jKf
I'd guess that the benchmarks which METR uses have enough label noise and other issues (e.g. specification ambiguity) that measuring >=95% reliability isn't meaningful. 80% probably is meaningful. When analyzing the high reliability regime, I think you'd want to get a ceiling on performance by baselining with human do...
2025-03-20T02:40:13.687Z
3
null
ZhDtfuBnxhC6X8ycG
Improved visualizations of METR Time Horizons paper.
improved-visualizations-of-metr-time-horizons-paper
https://www.lesswrong.com/posts/ZhDtfuBnxhC6X8ycG/improved-visualizations-of-metr-time-horizons-paper
LDJ
2025-03-19T23:36:52.771Z
a7mR7p2ZbHpgNY6Pq
> However, this strongly limits the space of possible aggregated agents. Imagine two EUMs, Alice and Bob, whose utilities are each linear in how much cake they have. Suppose they’re trying to form a new EUM whose utility function is a weighted average of their utility functions. Then they’d only have three options: > >...
2025-03-22T00:54:22.653Z
22
null
5tYTKX4pNpiG4vzYg
Towards a scale-free theory of intelligent agency
towards-a-scale-free-theory-of-intelligent-agency
https://www.mindthefuture.info/p/towards-a-scale-free-theory-of-intelligent
Richard_Ngo
2025-03-21T01:39:42.251Z
jos5sTh9KeDGZsZ5c
Actually, progress in 2024 is roughly 2x faster than earlier progress which seems consistent with thinking there is some distribution shift. It's just that this distribution shift didn't kick in until we had Anthropic competing with OpenAI and reasoning models. (Note that OpenAI didn't release a notably better model th...
2025-03-24T00:15:29.266Z
11
xCL4hw97qCEkDXtJR
deesrjitvXM4xYGZd
METR: Measuring AI Ability to Complete Long Tasks
metr-measuring-ai-ability-to-complete-long-tasks
https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/
Zach Stein-Perlman
2025-03-19T16:00:54.874Z
n2j2Nkgi4G5yzB6k7
My sense is that the GPT-2 and GPT-3 results are somewhat dubious, especially the GPT-2 result. It really depends on how you relate SWAA (small software engineering subtasks) to the rest of the tasks. My understanding is that no iteration was done though. However, note that it wouldn't be wildly more off trend if GPT-...
2025-03-24T00:21:44.644Z
12
QCdyw2pK4Mt39Locf
deesrjitvXM4xYGZd
METR: Measuring AI Ability to Complete Long Tasks
metr-measuring-ai-ability-to-complete-long-tasks
https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/
Zach Stein-Perlman
2025-03-19T16:00:54.874Z
AxXr5gpLBqd9fY6Dv
The list doesn't exclude Baumal effects as these are just the implication of: > - Physical bottlenecks and delays prevent growth. Intelligence only goes so far. > - Regulatory and social bottlenecks prevent growth this fast, INT only goes so far. Like Baumal effects are just some area of the economy with more limited...
2025-03-25T18:07:43.160Z
11
bsm8ZSPzoD4mYac9K
aD2RA3vtXs4p4r55b
On (Not) Feeling the AGI
on-not-feeling-the-agi
https://www.lesswrong.com/posts/aD2RA3vtXs4p4r55b/on-not-feeling-the-agi
Zvi
2025-03-25T14:30:02.215Z
i8JRxp4PpaRsun8tu
Sorry if my comment was triggering [@nostalgebraist](https://www.lesswrong.com/users/nostalgebraist?mention=user). : (
2025-03-26T03:29:40.275Z
2
AxXr5gpLBqd9fY6Dv
aD2RA3vtXs4p4r55b
On (Not) Feeling the AGI
on-not-feeling-the-agi
https://www.lesswrong.com/posts/aD2RA3vtXs4p4r55b/on-not-feeling-the-agi
Zvi
2025-03-25T14:30:02.215Z
tve3ipC9vbr3DbR2h
METR has found that substantially different scaffolding is most effective for o-series models. I get the sense that they weren't optimized for being effective multi-turn agents. At least, the o1 series wasn't optimized for this, I think o3 may have been.
2025-03-26T06:11:51.534Z
8
EAndFgfynLJ8FaoRq
4mvphwx5pdsZLMmpY
Recent AI model progress feels mostly like bullshit
recent-ai-model-progress-feels-mostly-like-bullshit
https://zeropath.com/blog/on-recent-ai-model-progress
lc
2025-03-24T19:28:43.450Z
3yNEed3p9TsGszNkg
Is this an accurate summary: - 3.5 substantially improved performance for your use case and 3.6 slightly improved performance. - The o-series models didn't improve performance on your task. (And presumably 3.7 didn't improve perf.) So, by "recent model progress feels mostly like bullshit" I think you basically just m...
2025-03-26T06:20:16.973Z
30
null
4mvphwx5pdsZLMmpY
Recent AI model progress feels mostly like bullshit
recent-ai-model-progress-feels-mostly-like-bullshit
https://zeropath.com/blog/on-recent-ai-model-progress
lc
2025-03-24T19:28:43.450Z
KexMzGJQ6eChztXpP
See also [Different senses in which two AIs can be "the same"](https://www.lesswrong.com/posts/4j6HJt8Exowmqp245/different-senses-in-which-two-ais-can-be-the-same).
2025-03-30T11:27:29.790Z
9
null
wQKskToGofs4osdJ3
The Pando Problem: Rethinking AI Individuality
the-pando-problem-rethinking-ai-individuality
https://www.lesswrong.com/posts/wQKskToGofs4osdJ3/the-pando-problem-rethinking-ai-individuality
Jan_Kulveit
2025-03-28T21:03:28.374Z
ydxoFix7bhnyTbzSE
Shouldn't a 32% increase in prices only make a modest difference to training FLOP? In particular, see the [compute forecast](https://ai-2027.com/research/compute-forecast). Between Dec 2026 and Dec 2027, compute increases by roughly an OOM and generally it looks like compute increases by a bit less than 1 OOM per year ...
2025-04-03T17:17:26.257Z
18
QjCdBBPnrkBJBAEFD
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
MCGcxRgZ4e5xzvrna
Of course, tariffs could have more complex effects than just reducing GPUs purchased by 32%, but this seems like a good first guess.
2025-04-03T17:18:00.418Z
6
ydxoFix7bhnyTbzSE
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
dyXpH4RtznCpQ727M
Sure, but note that the story "tariffs -> recession -> less AI investment" doesn't particularly depend on GPU tariffs!
2025-04-03T17:31:54.864Z
7
wnDJQLEGgjAfB9bke
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
xRf8CgJoz4ekYptfC
> I’ll define an “SIE” as “we can get >=5 OOMs of increase in effective training compute in <1 years without needing more hardware”. I This is as of the point of full AI R&D automation? Or as of any point?
2025-04-04T19:07:27.275Z
7
null
XDF6ovePBJf6hsxGj
Will compute bottlenecks prevent a software intelligence explosion?
will-compute-bottlenecks-prevent-a-software-intelligence-1
https://www.lesswrong.com/posts/XDF6ovePBJf6hsxGj/will-compute-bottlenecks-prevent-a-software-intelligence-1
Tom Davidson
2025-04-04T17:41:37.088Z
zAv7aQfWTWpzz23X2
I think it would be a mistake to interpret this paper as a substantial update against large safety gains from inspecting CoT. This paper exposes unfaithfulness in cases where the non-visible reasoning is extremely minimal such that it can easily happen within a forward pass (e.g. a simple reward hack or an easy-to-not...
2025-04-04T22:56:45.451Z
34
null
TmaahE9RznC8wm5zJ
AI CoT Reasoning Is Often Unfaithful
ai-cot-reasoning-is-often-unfaithful
https://www.lesswrong.com/posts/TmaahE9RznC8wm5zJ/ai-cot-reasoning-is-often-unfaithful
Zvi
2025-04-04T14:50:05.538Z
aFwbvo2xkycXBAWg5
> I'm skeptical regarding are the economic and practical implications (AGI labs' revenue tripling and 50% faster algorithmic progress) Notably, the trend in the last few years is that AI companies triple their revenue each year. So, the revenue tripling seems very plausible to me. As far as 50% algorithmic progress, ...
2025-04-07T16:30:25.781Z
6
g4sYfioiSWw9S9qJc
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
JyLo2QGGb73ciEEyP
I think the best source for revenue growth is [this](https://epoch.ai/data-insights/ai-companies-revenue) post from epoch. I think we only have the last 2 years really, (so "last few years" is maybe overstating it), but we do have revenue projections and we have more than 1 data point per year.
2025-04-07T18:32:55.990Z
5
pdJPZSfRsBobptAYB
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
wjCb34a7LWTR4xmR2
> A "no" to either would mean this work falls under milling behavior, and will not meaningfully contribute toward keeping humanity safe from DeepMind's own actions. I think it's probably possible greatly improve safety given a moderate budget for safety and not nearly enough buy in for (1) and (2). (At least not enoug...
2025-04-07T18:40:46.803Z
7
msnnJwRMdvwtBzCLi
3ki4mt4BA6eTx56Tc
Google DeepMind: An Approach to Technical AGI Safety and Security
google-deepmind-an-approach-to-technical-agi-safety-and
https://arxiv.org/abs/2504.01849
Rohin Shah
2025-04-05T22:00:14.803Z
eAjsrSrqxCiiHdG4v
I only skimmed this essay and I'm probably more sympathetic to moral patienthood of current AI systems than many, but I think this exact statement is pretty clearly wrong: > Statistically speaking, if you're an intelligent mind that came into existence in the past few years, you're probably running on a large language...
2025-04-07T22:36:30.512Z
1
null
m4ZpDyHQ2sz8FPDuN
Factory farming intelligent minds
factory-farming-intelligent-minds
https://www.lesswrong.com/posts/m4ZpDyHQ2sz8FPDuN/factory-farming-intelligent-minds
Odd anon
2025-04-07T20:05:04.064Z
7RQToQkKnRHrz55Y9
> the median narrative is probably around 2030 or 2031. (At least according to me. Eli Lifland is smarter than me and says December 2028, so idk.) Notably, this is Eli's forecast for "superhuman coder" which could be substantially before AIs are capable enough for takeover to be plausible. I think Eli's median for "A...
2025-04-09T22:25:06.824Z
5
null
Yzcb5mQ7iq4DFfXHx
Thoughts on AI 2027
thoughts-on-ai-2027
https://intelligence.org/2025/04/09/thoughts-on-ai-2027/
Max Harms
2025-04-09T21:26:23.926Z
MhwFxpEkawrx4bezb
I think it would be a mistake to interpret this paper as a substantial update against large safety gains from inspecting CoT. This paper exposes unfaithfulness in cases where the non-visible reasoning is extremely minimal such that it can easily happen within a forward pass (e.g. a simple reward hack or an easy-to-not...
2025-04-10T17:01:03.173Z
9
null
PrcBFPkoRNGWrvdPk
Reasoning models don't always say what they think
reasoning-models-don-t-always-say-what-they-think-1
https://www.anthropic.com/research/reasoning-models-dont-say-think
Joe Benton
2025-04-09T19:48:58.733Z
SzBMzQzKAXAmbXkzP
Are you sure that we see "vestigial reasoning" when: - We run a bunch of RL while aggressively trying to reduce CoT length (e.g., with a length penalty); - The input is in distribution with respect to the training distribution; - The RL is purely outcome based. I'd guess this mostly doesn't occur in this case and the...
2025-04-13T23:02:25.226Z
7
null
6AxCwm334ab9kDsQ5
Vestigial reasoning in RL
vestigial-reasoning-in-rl
https://www.lesswrong.com/posts/6AxCwm334ab9kDsQ5/vestigial-reasoning-in-rl
Caleb Biddulph
2025-04-13T15:40:11.954Z
bTakJ5ZSuvoyquruZ
My distribution is very uncertain, but I'd say 25% by June 2027 and 50% by Jan 2031. (I answer a similar question, but for a slightly higher bar of capabilties and operationalized somewhat differently [here](https://www.lesswrong.com/posts/FG54euEAesRkSZuJN/ryan_greenblatt-s-shortform?commentId=iwodobEWjt9qwHbb2). I'v...
2025-04-14T20:47:12.610Z
4
uYCcCaDEwXpaDeryg
gsj3TWdcBxwkm9eNt
≤10-year Timelines Remain Unlikely Despite DeepSeek and o3
10-year-timelines-remain-unlikely-despite-deepseek-and-o3
https://www.lesswrong.com/posts/gsj3TWdcBxwkm9eNt/10-year-timelines-remain-unlikely-despite-deepseek-and-o3
Rafael Harth
2025-02-13T19:21:35.392Z
bC5AiCjTS6AinFyXg
> This contradicts METR timelines, which, IMO, is the best piece of info we currently have for predicting when AGI will arrive. Have you read the [timelines supplement](https://ai-2027.com/research/timelines-forecast)? One of their main methodologies [involves using this exact data from METR](https://ai-2027.com/resea...
2025-04-19T21:36:15.326Z
6
xDkzLmzgK7K7uazFC
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
JgMRCYMr3p2iiKoE8
Looks like Eli beat me to the punch!
2025-04-19T21:38:32.112Z
6
bC5AiCjTS6AinFyXg
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
CAvWEwEvCuo8cGsZo
Isn't it kinda unreasonable to put 10% on superhuman coder in a year if current AIs have a 15 nanosecond time horizon? TBC, it seems fine IMO if the model just isn't very good at predicting the 10th/90th percentile, especially with extreme hyperparameters. ~~I also don't know how they ran this, I tried looking for mod...
2025-04-19T21:59:15.457Z
4
Dotw7krkeELvLP83p
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
7tzuoQbmsDnuNofjt
> I wouldn't be surprised if 3-5% of questions were mislabeled or impossible to answer, but 25-50%? You're basically saying that HLE is worthless. I'm curious why. Various people looked at randomly selected questions and found similar numbers. (I don't think the dataset is worthless, I think if you filtered down to t...
2025-04-19T22:15:06.092Z
9
7i4LAvKkzwwftWxpX
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
qz4iEse2S837qmPeX
It's notable that you're just generally arguing against having probabilistic beliefs about events which are unprecedented[^cont], nothing is specific to this case of doing AI forecasting. You're mostly objecting to the idea of having (e.g.) medians on events like this. [^cont]: Of course, the level of precedentedness ...
2025-04-21T01:06:27.442Z
4
vG9AJAkWX4Z7bFDtg
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
pd2ijtiHoovyvmgWH
(Yes, I'm aware you meant imprecise probabilities. These aren't probablities though (in the same sense that a range of numbers isn't a number), e.g., you're unwilling to state a median.)
2025-04-21T15:54:53.082Z
2
AokZ9hinwdJc28YGN
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
Jfh5BRoLGCZoAdKnc
Hmm, I think your argument is roughly right, but missing a key detail. In particular, the key aspect of the SARs (and higher levels of capability) is that they can be **strictly better than humans at everything** while simultaneously being 30x faster *and* 30x more numerous. (Or, there is 900x more parallel labor, but ...
2025-04-21T18:16:41.319Z
3
fzrqNyKgheH3mM4QG
bfHDoWLnBH9xR3YAK
AI 2027 is a Bet Against Amdahl's Law
ai-2027-is-a-bet-against-amdahl-s-law
https://www.lesswrong.com/posts/bfHDoWLnBH9xR3YAK/ai-2027-is-a-bet-against-amdahl-s-law
snewman
2025-04-21T03:09:40.751Z
TygofbgeG6g9d562j
> I note that I am confused by this diagram. In particular, the legend indicates a 90th percentile forecast of ">2100" for ASI, but the diagram appears to show the probability dropping to zero around the beginning of 2032. I think it's just that the tail is very long and flat with <1% per year. So, it looks like it go...
2025-04-21T18:24:14.182Z
7
null
bfHDoWLnBH9xR3YAK
AI 2027 is a Bet Against Amdahl's Law
ai-2027-is-a-bet-against-amdahl-s-law
https://www.lesswrong.com/posts/bfHDoWLnBH9xR3YAK/ai-2027-is-a-bet-against-amdahl-s-law
snewman
2025-04-21T03:09:40.751Z
z3tjK5oLFa5JkXtFX
Sure, but for output quality *better* than what humans could (ever) do to matter for the relative speed up, you have to argue about compute bottlenecks, not Amdahl's law for just the automation itself! (As in, if some humans would have done something in 10 years and it doesn't have any environmental bottleneck, then 1...
2025-04-21T18:37:08.279Z
2
2dx57dRaDrATYkHpL
bfHDoWLnBH9xR3YAK
AI 2027 is a Bet Against Amdahl's Law
ai-2027-is-a-bet-against-amdahl-s-law
https://www.lesswrong.com/posts/bfHDoWLnBH9xR3YAK/ai-2027-is-a-bet-against-amdahl-s-law
snewman
2025-04-21T03:09:40.751Z
9nckXCButbztEynJE
I'm worried that you're missing something important because you mostly argue against large AI R&D multipliers, but you don't spend much time directly referencing compute bottlenecks in your arguments that the forecast is too aggressive. Consider the case of doing pure math research (which we'll assume for simplicity d...
2025-04-21T18:47:06.538Z
15
null
bfHDoWLnBH9xR3YAK
AI 2027 is a Bet Against Amdahl's Law
ai-2027-is-a-bet-against-amdahl-s-law
https://www.lesswrong.com/posts/bfHDoWLnBH9xR3YAK/ai-2027-is-a-bet-against-amdahl-s-law
snewman
2025-04-21T03:09:40.751Z
r7mLTXGg36RK4eCNB
> I'm having trouble parsing this sentence You said "This is valid for activities which benefit from speed and scale. But when output quality is paramount, speed and scale may not always provide much help?". But, when considering activities that aren't bottlenecked on the environment, then to achieve 10x acceleration ...
2025-04-21T19:05:38.720Z
2
D2dy6FvahqEbG5DHy
bfHDoWLnBH9xR3YAK
AI 2027 is a Bet Against Amdahl's Law
ai-2027-is-a-bet-against-amdahl-s-law
https://www.lesswrong.com/posts/bfHDoWLnBH9xR3YAK/ai-2027-is-a-bet-against-amdahl-s-law
snewman
2025-04-21T03:09:40.751Z
iXCmtnpx7gZqziBjb
Another way to put this disagreement is that you can interpret all of the AI 2027 capability milestones as refering to the capability of the weakest bottlenecking capability, so: - Superhuman coder has to dominate *all* research engineers at *all* pure research engineering tasks. This includes the most bottlenecking c...
2025-04-21T19:15:51.958Z
9
fzrqNyKgheH3mM4QG
bfHDoWLnBH9xR3YAK
AI 2027 is a Bet Against Amdahl's Law
ai-2027-is-a-bet-against-amdahl-s-law
https://www.lesswrong.com/posts/bfHDoWLnBH9xR3YAK/ai-2027-is-a-bet-against-amdahl-s-law
snewman
2025-04-21T03:09:40.751Z
ame49sHq5BXfgbSEn
> This is both projected forward and treated with either 1 (in 45% of cases) or 2 (in all cases) super-exponential terms that make it significantly faster than an inferred 4.6x per year. Hmm, I think you're looking at the [more basic trend extrapolation for the timelines model](https://ai-2027.com/research/timelines-f...
2025-04-21T23:01:02.094Z
6
62uRL6KFZytcrdRGL
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
Ahd4tCfF2zFRsS9tp
> zero-shot WikiText103 perplexity and 5-shot MMLU These are somewhat awkward benchmarks because they don't actually measure downstream usefulness at software engineering or AI research. In particular, these tasks might not measure improvements in RL which can have huge effects on usefulness and have seen fast algorit...
2025-04-21T23:09:51.312Z
13
62uRL6KFZytcrdRGL
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
QNcgmdLNk78fAatCq
> In all timelines models presented there is acknowledgement that compute does not accelerate. When you say "accelerate" do you mean "the rate of compute scaling increases"? I agree they aren't expecting this (and roughly model a fixed rate of compute progress which matches historical trends as this is just an extrapo...
2025-04-21T23:16:03.536Z
4
oAoF9zuBmruqdcTNn
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
FhujzkmBNbPyGAMve
> I said there was no compute acceleration not that there was no more compute scaling? (Yes, sorry, edited my original comment to clarify.) > a equal and super-accelerated algorithmic or efficiency term (v_algorithmic) as shown in the code here I don't think the "AI assisted AI R&D" speed ups along the way to super...
2025-04-21T23:20:50.137Z
4
kHtiJNBBujKu6BdH5
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
rrzRWKhEMHv8hqT88
Sure. [Epoch estimates](https://epoch.ai/data/notable-ai-models?view=table) 2e25 flop for GPT-4 and 3.4e24 for deepseek V3. So a bit less than 10x actually, but quite close. (And V3 is substantially better.) R1 is [around 1/6 of deepseek V3 cost](https://epoch.ai/gradient-updates/what-went-into-training-deepseek-r1).
2025-04-21T23:25:03.990Z
4
EMCKmSyrvpNkBztHX
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
7qF5o4zfeHfKipeTL
For context, my preregistered guess would be that AI R&D speed ups along the way to superhuman coder make it come around 1.5x faster, though between 1.25-2 all are consistent with my best guess. (So e.g., rather than ~2029.0 median on Eli's model without intermediate AI R&D speed ups we'd see around 2031.0 or so. I'd e...
2025-04-21T23:28:47.276Z
4
aqbiCR4ZbYDw67Y2w
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
G8uD8x5vybYDJkutx
Just ran the code and it looks like I'm spot on and the median goes to Mar 2031.
2025-04-21T23:32:41.572Z
2
7qF5o4zfeHfKipeTL
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
dAwRr9dFe9br2NXkK
> it does not meet the standard. Importantly, it is much better than GPT-4 on the relevant downstream tasks.
2025-04-22T00:09:11.335Z
1
pfnQFjececmFCE86g
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
BB7u5vJixhdYdXxH7
> R1 can't possibly be below V3 cost because it is inclusive? If I'm not mistaken, R1 is not trained from scratch, but I could be wrong. Yes, I meant 1/6 additional cost which is ~negligable.
2025-04-22T00:10:20.795Z
1
pfnQFjececmFCE86g
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
x2gqgjzajDrNGCkKn
> Please let me know if you suspect I've over-interpreted that validation. Slightly? My view is more like: - For AIs to be superhuman AI researchers, they probably need to match humans at most underlying/fundamental cognitive tasks, including reasonably sample efficient learning. (Or at least learning which is compet...
2025-04-22T00:37:11.063Z
5
DsLAJgwsfiBsW9pKe
bfHDoWLnBH9xR3YAK
AI 2027 is a Bet Against Amdahl's Law
ai-2027-is-a-bet-against-amdahl-s-law
https://www.lesswrong.com/posts/bfHDoWLnBH9xR3YAK/ai-2027-is-a-bet-against-amdahl-s-law
snewman
2025-04-21T03:09:40.751Z
Sk9DibiWyRuBypNGR
> If it is compute efficient according to even Kaplan or Chinchilla scaling laws, please demonstrate that for me. We only have [leaked numbers confirming reasonably efficient training](https://semianalysis.com/2023/07/10/gpt-4-architecture-infrastructure/) but GPT-4 is widely believed to be a quite efficient model for...
2025-04-22T00:48:40.468Z
1
3YP9LjWfLRAgB9SY2
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
TSRjhsmctLSRiKTtg
("Has to" is maybe a bit strong, I think I probably should have said "will probably end up needing to be better competitive with the best human experts at basically everything (other than vision) and better at more central AI R&D given the realistic capability profile". I think I generally expect full automation to hit...
2025-04-22T03:11:53.632Z
3
qZJgyF2cdMZ2JsHq7
bfHDoWLnBH9xR3YAK
AI 2027 is a Bet Against Amdahl's Law
ai-2027-is-a-bet-against-amdahl-s-law
https://www.lesswrong.com/posts/bfHDoWLnBH9xR3YAK/ai-2027-is-a-bet-against-amdahl-s-law
snewman
2025-04-21T03:09:40.751Z
Y9cQMZ6Yse4oAt74u
> Maybe a crux here is maybe how big the speedup is? What you describe are good reasons why companies are unlikely to want to release this information unilaterially, but from a safety perspective, we should instead consider how imposing such a policy alters the overall landscape. From this perspective, the main quest...
2025-04-22T15:11:04.944Z
4
Thbhg8TFBXim3WRjc
FGqfdJmB8MSH5LKGc
Training AGI in Secret would be Unsafe and Unethical
training-agi-in-secret-would-be-unsafe-and-unethical-1
https://www.lesswrong.com/posts/FGqfdJmB8MSH5LKGc/training-agi-in-secret-would-be-unsafe-and-unethical-1
Daniel Kokotajlo
2025-04-18T12:27:35.795Z
XsSFj5qG4HozF6RbM
I think my description is consistent with "some activities on which the SAR is worse" as long as these aren't bottlenecking and it is overall dominating human researchers (as in, adding human researchers is negligable value). But whatever, you're the author here. Maybe "Superhuman coder has to dominate all research e...
2025-04-22T17:05:32.632Z
2
fttRA6GCGZL476orB
bfHDoWLnBH9xR3YAK
AI 2027 is a Bet Against Amdahl's Law
ai-2027-is-a-bet-against-amdahl-s-law
https://www.lesswrong.com/posts/bfHDoWLnBH9xR3YAK/ai-2027-is-a-bet-against-amdahl-s-law
snewman
2025-04-21T03:09:40.751Z
EAtYrfAiks2EifQpx
I think o3 maybe does worse on RE-bench than 3.7 sonnet due to often attempting reward hacking. It could also be noise, it is just a small number of tasks. (Presumably these reward hacks would have worked better in the OpenAI training setup but METR filters it out?) It doesn't attempt reward hacking as much / as aggres...
2025-04-24T18:07:52.430Z
5
htTyS4c7N3kwjsjbZ
TpSFoqoG2M5MAAesg
AI 2027: What Superintelligence Looks Like
ai-2027-what-superintelligence-looks-like-1
https://ai-2027.com/
Daniel Kokotajlo
2025-04-03T16:23:44.619Z
GFthzNEvfuecwqL4u
I buy your arguments for optimism about not needing to simplify/change our goals to compete. (I also think that there are other stronger reasons to expect we don't need goal simplification like just keeping humans alive and later giving back the resources which is quite simple and indirectly points to what humans want....
2025-04-24T18:16:33.444Z
2
nigkBt47pLMi5tnGd
5tYTKX4pNpiG4vzYg
Towards a scale-free theory of intelligent agency
towards-a-scale-free-theory-of-intelligent-agency
https://www.mindthefuture.info/p/towards-a-scale-free-theory-of-intelligent
Richard_Ngo
2025-03-21T01:39:42.251Z
43Q72wsTwvojLGExz
This is one of my favorite current research directions.
2025-04-25T00:15:14.049Z
5
null
ARQs7KYY9vJHeYsGc
Modifying LLM Beliefs with Synthetic Document Finetuning
modifying-llm-beliefs-with-synthetic-document-finetuning
https://alignment.anthropic.com/2025/modifying-beliefs-via-sdf/
RowanWang
2025-04-24T21:15:17.366Z
xhB8JRFistvm2ehQQ
Some reasons for this (that I quickly wrote in response to someone asking a question about this): - There aren't that many research direction we can do now which plausibly transfer to later much more powerful AIs while if we got really good at this it could transfer. (Up to around or a bit beyond full AI R&D automatio...
2025-04-25T00:16:35.955Z
16
43Q72wsTwvojLGExz
ARQs7KYY9vJHeYsGc
Modifying LLM Beliefs with Synthetic Document Finetuning
modifying-llm-beliefs-with-synthetic-document-finetuning
https://alignment.anthropic.com/2025/modifying-beliefs-via-sdf/
RowanWang
2025-04-24T21:15:17.366Z
AJHftvL6RWM6uZqdb
And there is some ongoing future work which is currently trying to meet this higher standard! (Or at least something similar to it.)
2025-04-25T16:17:05.234Z
5
jXKPs7bu7XGfbmbB9
ARQs7KYY9vJHeYsGc
Modifying LLM Beliefs with Synthetic Document Finetuning
modifying-llm-beliefs-with-synthetic-document-finetuning
https://alignment.anthropic.com/2025/modifying-beliefs-via-sdf/
RowanWang
2025-04-24T21:15:17.366Z
sZKbdhdrQ6HRLyXZe
> When AI companies have human-level AI systems, will they use them for alignment research, or will they use them (mostly) to advance capabilities instead? It's not clear this is a crux for the automating alignment research plan to work out. In particular, suppose an AI company currently spends 5% of its resources on...
2025-04-25T19:33:41.458Z
4
null
XLNxrFfkyrdktuzqn
Why would AI companies use human-level AI to do alignment research?
why-would-ai-companies-use-human-level-ai-to-do-alignment
https://www.lesswrong.com/posts/XLNxrFfkyrdktuzqn/why-would-ai-companies-use-human-level-ai-to-do-alignment
MichaelDickens
2025-04-25T19:12:56.202Z
DYb9tmHLHrewfqnNi
People interested in approaches to solving reward hacking which don't depend on supervision (and thus might scale to superintelligence) should consider looking at [our earlier work on measurement tampering](https://www.alignmentforum.org/posts/inALbAqdx63KTaGgs/benchmarks-for-measurement-tampering-detection). I don't p...
2025-04-27T02:21:54.720Z
4
jh7RMbXDfyzgum6ij
rKC4xJFkxm6cNq4i9
Reward hacking is becoming more sophisticated and deliberate in frontier LLMs
reward-hacking-is-becoming-more-sophisticated-and-deliberate
https://www.lesswrong.com/posts/rKC4xJFkxm6cNq4i9/reward-hacking-is-becoming-more-sophisticated-and-deliberate
Kei Nishimura-Gasparian
2025-04-24T16:03:57.359Z
KmhfBChzzsT7WmyBi
Ironically, arguably the most important/useful point of the essay is arguing for a rebranded version of the "precisely timed short slow/pause/pivot resources to safety" proposal. Dario's rebranded it as spending down a "security buffer". (I don't have a strong view on whether this is a good rebrand, seems reasonable t...
2025-04-28T02:16:08.406Z
8
n5fJLWCGBSDApESM9
SebmGh9HYdd8GZtHA
"The Urgency of Interpretability" (Dario Amodei)
the-urgency-of-interpretability-dario-amodei
https://www.darioamodei.com/post/the-urgency-of-interpretability
RobertM
2025-04-27T04:31:50.090Z
fFKDdzNiQyYpdBfGi
Another potential crux[^related] is that Ege's world view seemingly doesn't depend at all on AIs which are much faster and smarter than any human. As far as I can tell, it doesn't enter into his modeling of takeoff (or timelines to full automation of remote work which partially depends on something more like takeoff). ...
2025-04-28T02:52:47.776Z
20
null
xxxK9HTBNJvBY2RJL
The case for multi-decade AI timelines [Linkpost]
the-case-for-multi-decade-ai-timelines-linkpost
https://epoch.ai/gradient-updates/the-case-for-multi-decade-ai-timelines
Noosphere89
2025-04-27T15:31:47.902Z
4arc3J2Z3G68rmrpi
I think your discussion (and Epoch's discussion) of the CES model is confused as you aren't taking into account the possibility that we're *already* bottlenecking on compute or labor. That is, I think you're making some assumption about the current marginal returns which is non-obvious and, more strongly, would be an a...
2025-04-28T07:21:13.434Z
83
null
XDF6ovePBJf6hsxGj
Will compute bottlenecks prevent a software intelligence explosion?
will-compute-bottlenecks-prevent-a-software-intelligence-1
https://www.lesswrong.com/posts/XDF6ovePBJf6hsxGj/will-compute-bottlenecks-prevent-a-software-intelligence-1
Tom Davidson
2025-04-04T17:41:37.088Z
59NufM2Nkiux6QybX
As far as I can tell, this sort of consideration is at least somewhat damning for the literal CES model (with poor substitution) in any situation where the inputs have varied by hugely different amounts (many orders of magnitude of difference like in the compute vs labor case) and relative demand remains roughly simila...
2025-04-28T07:32:42.197Z
20
4arc3J2Z3G68rmrpi
XDF6ovePBJf6hsxGj
Will compute bottlenecks prevent a software intelligence explosion?
will-compute-bottlenecks-prevent-a-software-intelligence-1
https://www.lesswrong.com/posts/XDF6ovePBJf6hsxGj/will-compute-bottlenecks-prevent-a-software-intelligence-1
Tom Davidson
2025-04-04T17:41:37.088Z
noxcihBEW3FGTFC43
> I don’t see the trends that one would extrapolate in order to arrive at very short timelines on the order of a few years. The obvious trend extrapolations for AI’s economic impact give timelines to full remote work automation of around a decade, and I expect these trends to slow down by default. Actually, I think t...
2025-04-28T23:50:44.998Z
8
null
xxxK9HTBNJvBY2RJL
The case for multi-decade AI timelines [Linkpost]
the-case-for-multi-decade-ai-timelines-linkpost
https://epoch.ai/gradient-updates/the-case-for-multi-decade-ai-timelines
Noosphere89
2025-04-27T15:31:47.902Z
unWYWWyNMfoF8Bybd
> If we adjust for the 5-18x speed improvement measured for experienced workers, and target an 80% task success rate, that pushes the timeline out by over three years I don't think this is a good interpretation of the 5-18x multiplier. In particular, I think the "acquire context multiplier" will be increasingly small ...
2025-04-30T05:39:32.273Z
15
null
fRiqwFPiaasKxtJuZ
Interpreting the METR Time Horizons Post
interpreting-the-metr-time-horizons-post
https://amistrongeryet.substack.com/p/measuring-ai-progress
snewman
2025-04-30T03:03:19.928Z
JieqtgAspBYFh3CCf
> In actuality, the study doesn’t say much about AGI, except to provide evidence against the most aggressive forecasts. This feels quite wrong to me. Surely if AIs were completing 1 month long self contained software engineering tasks (e.g. what a smart intern might do in the first month) that would be a big update to...
2025-04-30T05:51:22.817Z
5
null
fRiqwFPiaasKxtJuZ
Interpreting the METR Time Horizons Post
interpreting-the-metr-time-horizons-post
https://amistrongeryet.substack.com/p/measuring-ai-progress
snewman
2025-04-30T03:03:19.928Z
tT2qxAwDfsA5c9jxp
Personally, I updated toward shorter timelines upon seeing a preliminary version of their results which just showed the more recent doubling trend and then updated most of the way back on seeing the longer run trend. (Or maybe even toward slightly longer timelines than I started with, I forget.)
2025-04-30T05:53:42.291Z
3
JieqtgAspBYFh3CCf
fRiqwFPiaasKxtJuZ
Interpreting the METR Time Horizons Post
interpreting-the-metr-time-horizons-post
https://amistrongeryet.substack.com/p/measuring-ai-progress
snewman
2025-04-30T03:03:19.928Z
qy8BGNeJjwKTcJ7my
To be clear, I agree it provides evidence against very aggressive timelines (if I had 2027 medians I would have updated to longer), I was disagreeing with "the study doesn’t say much about AGI, except to". I think the study does provide a bunch of evidence about when AGI might come! (And it seems you agree.) I edited m...
2025-04-30T16:03:24.717Z
3
EgJQpJkRBBheZyGqp
fRiqwFPiaasKxtJuZ
Interpreting the METR Time Horizons Post
interpreting-the-metr-time-horizons-post
https://amistrongeryet.substack.com/p/measuring-ai-progress
snewman
2025-04-30T03:03:19.928Z