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
dn69tDosnhkewBKAi
> If I'm understanding the claims right, it seems like it'd be super crazy to bit the bullet? If you don't think human speed impacts the rate of technological progress, then what does? Literal calendar time? What would be the mechanism for that? Physical bottlenecks, compute bottlenecks, etc. The claim that you can o...
2025-05-01T02:09:14.620Z
5
KxGgoNuWhckuXFCDN
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
Y7Xt4pnG47DiZHKFi
> When we say that we hope that “whenever we observe a phenomenon in a neural net we can find an adequate explanation for why that phenomenon occurs”, that can't be quite right the way it’s stated. After all, some phenomena simply don’t have any real explanation. Maybe I'm confused, but can't we just: 1. Require that...
2025-05-01T02:36:12.149Z
3
null
xtcpEceyEjGqBCHyK
Obstacles in ARC's agenda: Finding explanations
obstacles-in-arc-s-agenda-finding-explanations
https://www.lesswrong.com/posts/xtcpEceyEjGqBCHyK/obstacles-in-arc-s-agenda-finding-explanations
David Matolcsi
2025-04-30T23:03:31.327Z
gLtFb534fMkgFPadQ
You might hope for elicitation efficiency, as in, you heavily RL the model to produce useful considerations and hope that your optimization is good enough that it covers everything well enough. Or, two lower bars you might hope for: - It brings up considerations that it "knows" about. (By "knows" I mean relatively de...
2025-05-01T03:15:24.772Z
2
GTY4TMES97quHscaR
nJcuj4rtuefeTRFHp
Can we safely automate alignment research?
can-we-safely-automate-alignment-research
https://joecarlsmith.com/2025/04/30/can-we-safely-automate-alignment-research
Joe Carlsmith
2025-04-30T17:37:13.193Z
CRYAbJS6seyZatdf7
Yeah, I agree with this and doesn't seem that productive to speculate about people's views when I don't fully understand them. > They might say something like: "We're only asserting that labor and compute are complementary. That means it's totally possible that slowing down humans would slow progress a lot, but that s...
2025-05-01T17:59:26.008Z
4
8eeBJCjdfCHmKqwQv
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
hYtHj3DCFg6JLWcJM
Gotcha. I agree with 1-4, but I'm not sure I agree with 5, at least I don't agree that 5 is separate from 4. In particular: If we can make an explanation for some AI while we're training it and this is actually a small increase in cost, then we can apply this to input distribution generator. This doesn't make trainin...
2025-05-01T20:59:32.862Z
3
SuFaAZhwSoMAZNQFX
xtcpEceyEjGqBCHyK
Obstacles in ARC's agenda: Finding explanations
obstacles-in-arc-s-agenda-finding-explanations
https://www.lesswrong.com/posts/xtcpEceyEjGqBCHyK/obstacles-in-arc-s-agenda-finding-explanations
David Matolcsi
2025-04-30T23:03:31.327Z
CqXQMfWx392zw2Xyz
Right now, the cost of feeding all humans is around 1% of GDP. It's even cheaper to keep people alive for a year as the food is already there and converting this food into energy for AIs will be harder than getting energy other ways. If GDP has massively increased due to powerful AIs, the relative cost would go down f...
2025-05-01T23:32:58.248Z
5
Bri4JynKhAN4PBuFn
pZhEQieM9otKXhxmd
Gradual Disempowerment: Systemic Existential Risks from Incremental AI Development
gradual-disempowerment-systemic-existential-risks-from
https://gradual-disempowerment.ai/
Jan_Kulveit
2025-01-30T17:03:45.545Z
tH8wQwvSMJvcH7Rce
Thanks for the post, the discussion about [compute scaling slowing](https://www.lesswrong.com/posts/XiMRyQcEyKCryST8T/slowdown-after-2028-compute-rlvr-uncertainty-moe-data-wall#Training_Compute_Slowdown) seems right to me.[^marginsdecrease] [^marginsdecrease]: That said, I think we might see atypically high improvemen...
2025-05-02T18:05:04.889Z
17
null
XiMRyQcEyKCryST8T
Slowdown After 2028: Compute, RLVR Uncertainty, MoE Data Wall
slowdown-after-2028-compute-rlvr-uncertainty-moe-data-wall
https://www.lesswrong.com/posts/XiMRyQcEyKCryST8T/slowdown-after-2028-compute-rlvr-uncertainty-moe-data-wall
Vladimir_Nesov
2025-05-01T13:54:49.446Z
8hZ6KAenPjQetYuL8
Sure, but none of these things are cruxes for the argument I was making which was that it wasn't that expensive to keep humans physically alive. I'm not denying that humans might all be out of work quickly (putting aside regulatory capture, goverment jobs, human job programs, etc). My view is more that if alignment is...
2025-05-02T18:42:13.663Z
5
S2GiKYy5yLo7oopSB
pZhEQieM9otKXhxmd
Gradual Disempowerment: Systemic Existential Risks from Incremental AI Development
gradual-disempowerment-systemic-existential-risks-from
https://gradual-disempowerment.ai/
Jan_Kulveit
2025-01-30T17:03:45.545Z
esbXheWAup8wXfk7p
> Intuitively, I still have trouble accepting the very high speedup factors contemplated in the AI 2027 model. This could be a failure of my intuition. Curious how you feel about the following intuition pump: Imagine an AI company where all of their researchers are 100x slower (as in, they think and write code at 100...
2025-05-03T00:45:03.247Z
8
null
FFKnWk2MGJmyQWEsd
Updates from Comments on "AI 2027 is a Bet Against Amdahl's Law"
updates-from-comments-on-ai-2027-is-a-bet-against-amdahl-s
https://www.lesswrong.com/posts/FFKnWk2MGJmyQWEsd/updates-from-comments-on-ai-2027-is-a-bet-against-amdahl-s
snewman
2025-05-02T23:52:29.282Z
CTmx3tWfAqaA7PpNK
Nitpick: > The average Mechanical Turker gets a little over 75%, far less than o3’s 87.5%. Actually, average Mechanical Turk performance is closer to 64% on the ARC-AGI evaluation set. Source: [https://arxiv.org/abs/2409.01374](https://arxiv.org/abs/2409.01374). (Average performance on the training set is around 76%...
2025-05-04T00:50:24.203Z
5
null
R7r8Zz3uRyjeaZbss
"Superhuman" Isn't Well Specified
superhuman-isn-t-well-specified
https://justismills.substack.com/p/superhuman-isnt-well-specified
JustisMills
2025-05-03T23:42:11.964Z
eJ59ck4o57b8KyYDq
Should be possible for agents with long run preferences to [strategy steal](https://www.lesswrong.com/posts/nRAMpjnb6Z4Qv3imF/the-strategy-stealing-assumption), so I don't see why evolution is an issue from this perspective.
2025-05-04T04:34:40.116Z
2
jEfSZFNaimTpajdfo
fMqgLGoeZFFQqAGyC
How do we solve the alignment problem?
how-do-we-solve-the-alignment-problem
https://joecarlsmith.substack.com/p/how-do-we-solve-the-alignment-problem
Joe Carlsmith
2025-02-13T18:27:27.712Z
dHtqraiAmfGqWZzzb
IMO, works much better to use SOTA LLMs over google translate, at least last I checked.
2025-05-04T04:40:32.244Z
6
pFhscXeaATR9spktp
L9Cf6ZHkhJYX7afff
Does translating a post with an LLM affect its rating?
does-translating-a-post-with-an-llm-affect-its-rating
https://www.lesswrong.com/posts/L9Cf6ZHkhJYX7afff/does-translating-a-post-with-an-llm-affect-its-rating
ReverendBayes
2025-05-03T14:45:01.070Z
RSsM6pATtJxpWTvRB
I don't think it is specific to pixel art, I think it is more about general visual understanding, particularly when you have to figure out downstream consequences from the visual understanding (like "walk to here").
2025-05-04T04:42:54.627Z
4
hdesNcKJyQa9YuGKk
EjgTgooax7oypqdS2
What's up with AI's vision
what-s-up-with-ai-s-vision
https://www.lesswrong.com/posts/EjgTgooax7oypqdS2/what-s-up-with-ai-s-vision
Joachim Bartosik
2025-05-03T13:23:42.631Z
xHqmGvshpgCk4oLYR
Does mundane mandy care about stuff outside the solar system? Let alone stuff which is over 1 million light years away. (Separately, I think the distal light cone is more like 10 B ly than 45 B ly as we can only reach a subset of the observable universe.)
2025-05-04T20:41:46.048Z
5
Fkw5Ks82HxGHxsbK5
MDgEfWPrvZdmPZwxf
Why I am not a successionist
why-i-am-not-a-successionist
https://ninapanickssery.substack.com/p/aligned-ai-should-not-replace-humanity
Nina Panickssery
2025-05-04T19:08:52.645Z
A64xbyQv2BsZ7yvvh
Suppose your view was that P(AGI if no pause/slow before 2050) = 80%. Then, if we condition on AGI after 2050, surely most of the probability mass isn't due to pausing/slowing right? So, what would be the mechanism if not some sort of technical research or exogenous factor (e.g. society getting wiser) over the interve...
2025-05-05T22:47:12.389Z
3
zz6CtnyaNBDcjeTsu
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
poBdGeLhBQdAbJmg5
> I expect the general chaos combined with supply chain disruptions around Taiwan to have slowed things down. Won't supply chain disruption around Taiwan take at least many months to cause a considerable slowdown? Naively, we might expect that AI companies get around 10% more FLOP/s each month (for around 3.5x per ye...
2025-05-05T22:59:59.561Z
3
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
S6Fpcs9WbKnN2FxSa
Not that important to get into, but I'd guess the probability of >3 decade long coordinated pause prior to "very scary AI (whatever you were worried might takeover)" (which is maybe better to talk about than AGI) is like 3% and I'm sympathetic to lower. I'm skeptical of 95% on P(AGI < 2050|no pause) unless you mean so...
2025-05-06T18:16:20.862Z
7
SefgSJfQNgRxpRwxx
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
jtHGsT58kbsyauMmq
> Noting one other dynamic: advanced models are probably not going to act misaligned in everyday use cases (that consumers have an incentive to care about, though again revealed preference is less clear), even if they're misaligned. That's the whole deceptive alignment thing. Agreed, but customers would also presumabl...
2025-05-06T18:34:30.881Z
3
Eu3vwpb3zd3uLQbZu
wBTNkfukMsmjtgcnW
Should there be just one western AGI project?
should-there-be-just-one-western-agi-project
https://www.forethought.org/research/should-there-be-just-one-western-agi-project
rosehadshar
2024-12-03T10:11:17.914Z
od5k3TZ3PkD2CgvoY
I don't think Metaculus is that confident. Some questions: - https://www.metaculus.com/questions/19356/transformative-ai-date/ - https://www.metaculus.com/questions/5406/world-output-doubles-in-4-years-by-2050/ - https://www.metaculus.com/questions/5121/date-of-artificial-general-intelligence/ (the resolution criteria...
2025-05-06T20:44:25.621Z
3
W7sBkfn9qnug6Lvyp
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
eCL8q5je6e6fdx8Lc
> I generally don't think it's a good idea to put a probability on things where you have a significant ability to decide the outcome (i.e. probability of getting divorced), and instead encourage you to believe in pausing. In this case, I can at least talk about the probability of a multi decade pause (with the motivat...
2025-05-06T20:46:39.182Z
2
W7sBkfn9qnug6Lvyp
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
8MhKZQeyMqvNejmFs
I think this post would be better if it taboo'd the word alignment or at least defined it. I don't understand what the post means by alignment. My best guess is "generally being nice", but I don't see why this is what we wanted. I usually use the term alignment to refer to alignment between the AI and the developer, o...
2025-05-07T14:41:39.161Z
17
null
sDuWXb8cPXZ2yHdH4
Alignment first, intelligence later
alignment-first-intelligence-later
https://chrislakin.blog/p/alignment-first-intelligence-later
Chris Lakin
2025-03-30T22:26:55.302Z
LqLAgcJeAqwzELhth
Met in person or have other private knowledge also seems reasonable IMO.
2025-05-08T00:35:27.059Z
2
zfRNJkGxgNwovsZcG
LvYq2FEkPgoRLXgNd
Which journalists would you give quotes to? [one journalist per comment, agree vote for trustworthy]
which-journalists-would-you-give-quotes-to-one-journalist
https://www.lesswrong.com/posts/LvYq2FEkPgoRLXgNd/which-journalists-would-you-give-quotes-to-one-journalist
Nathan Young
2025-05-07T18:39:24.322Z
pBFfARipFYnJFB4oW
I agree about reflexive endorsement being important, at least eventually, but don't think this is out of reach while still having robust spec compliance and corrigibility.[^complex] [^complex]: Humans often endorse complex or myopic drives on reflection! This isn't something which is totally out of reach. Probably no...
2025-05-08T01:30:29.198Z
3
uYnh2k2X3TDGqgeEq
sDuWXb8cPXZ2yHdH4
Alignment first, intelligence later
alignment-first-intelligence-later
https://chrislakin.blog/p/alignment-first-intelligence-later
Chris Lakin
2025-03-30T22:26:55.302Z
ASRdoxRAMo9CCmWXf
> Doesn't seem that wild to me? When we scale up compute we're also scaling up the size of frontier training runs; maybe past a certain point running smaller experiments just isn't useful (e.g. you can't learn anything from experiments using 1 billionth of the compute of a frontier training run); and maybe past a certa...
2025-05-08T16:08:21.660Z
2
43pLgvubcdQ5arLmT
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
zya7SYT8wc7idFdjF
While I think debate might be a useful prosaic method for some levels of capability, I have a variety of concerns about this approach resulting in worst case guarantees[^call]: [^call]: I've also discussed these concerns in a call where the authors were present, but I thought it might be helpful to quickly write up my...
2025-05-14T18:27:44.091Z
20
null
iELyAqizJkizBQbfr
An alignment safety case sketch based on debate
an-alignment-safety-case-sketch-based-on-debate
https://arxiv.org/abs/2505.03989
Marie_DB
2025-05-08T15:02:06.345Z
3b6TkyAeiKxxnHTKb
> It is extremely difficult to gracefully put your finger on the scale of an LLM, to cause it to give answers it doesn’t ‘want’ to be giving. You will be caught. IMO, this takeaway feels way too strong. It could just be that this wasn't a very competent attempt. (And based on the system prompt we've seen, it sure look...
2025-05-16T16:57:22.804Z
53
null
kMH8zFoHHJvy6wH7h
Regarding South Africa
regarding-south-africa
https://www.lesswrong.com/posts/kMH8zFoHHJvy6wH7h/regarding-south-africa
Zvi
2025-05-16T16:10:07.989Z
iBixRcJDZnvoEZC2F
The key question is whether you can find improvements which work at large scale using mostly small experiments, not whether the improvements work just as well at small scale. The 3 largest algorithmic advances discussed here (Transformer, MoE, and MQA) were all originally found at tiny scale (~1 hr on an H100 or ~1e19...
2025-05-17T22:43:15.572Z
98
null
qhjNejRxbMGQp4wHt
How Fast Can Algorithms Advance Capabilities? | Epoch Gradient Update
how-fast-can-algorithms-advance-capabilities-or-epoch
https://epoch.ai/gradient-updates/how-fast-can-algorithms-advance-capabilities
henryj
2025-05-16T21:38:46.822Z
MbpoZcnutELxBcTPW
I'm a bit confused by what's going on with the paper claiming their empirical results support innovations being compute-dependent when they only test MQA (and IMO show unclear results in this case). It almost seems like they forgot to include results or didn't realize they only tested MQA because (e.g.) they talk about...
2025-05-18T03:41:19.830Z
9
iBixRcJDZnvoEZC2F
qhjNejRxbMGQp4wHt
How Fast Can Algorithms Advance Capabilities? | Epoch Gradient Update
how-fast-can-algorithms-advance-capabilities-or-epoch
https://epoch.ai/gradient-updates/how-fast-can-algorithms-advance-capabilities
henryj
2025-05-16T21:38:46.822Z
bw5Fh3ERCaczwyFnC
To be clear, I agree that reducing availability of compute will substantially slow algorithmic research. So, export controls which do a good job of reducing the available amount of compute would slow algorithmic research progress. If we have a fixed quantity (and quality) of human researchers and reduce the amount of ...
2025-05-18T17:42:42.436Z
6
iBixRcJDZnvoEZC2F
qhjNejRxbMGQp4wHt
How Fast Can Algorithms Advance Capabilities? | Epoch Gradient Update
how-fast-can-algorithms-advance-capabilities-or-epoch
https://epoch.ai/gradient-updates/how-fast-can-algorithms-advance-capabilities
henryj
2025-05-16T21:38:46.822Z
CNpcfrKBdziC3L56s
I bet o3 is better than 4.1. I also bet reasoning probably helps some.
2025-05-19T00:46:28.771Z
3
dvwaGJK2JR3iziY2q
xMGmibZpPDnawjHXk
Generating the Funniest Joke with RL (according to GPT-4.1)
generating-the-funniest-joke-with-rl-according-to-gpt-4-1
https://www.lesswrong.com/posts/xMGmibZpPDnawjHXk/generating-the-funniest-joke-with-rl-according-to-gpt-4-1
agg
2025-05-16T05:09:56.120Z
gvzL6ABvuSZajGr8E
(nit: global nuclear war isn't an existential catastrophe.)
2025-05-20T20:34:37.568Z
7
null
AiqDepus9e7Ty8mbD
Outcomes of the Geopolitical Singularity
outcomes-of-the-geopolitical-singularity
https://www.lesswrong.com/posts/AiqDepus9e7Ty8mbD/outcomes-of-the-geopolitical-singularity
Nikola Jurkovic
2025-05-20T18:09:43.494Z
urfb6gQKZnMENioC5
My favorite approach for the exploitable search problem is to try to minimize KL relative to human imitation. More precisely, consider any token prefix part way though a trajectory from an AI. We'll then have a human continue this trajectory (as effectively and safely as they can) and then KL regularize relative to thi...
2025-05-21T18:27:27.486Z
2
null
CuneN5HmLnztsLRzD
Unexploitable search: blocking malicious use of free parameters
unexploitable-search-blocking-malicious-use-of-free-1
https://www.lesswrong.com/posts/CuneN5HmLnztsLRzD/unexploitable-search-blocking-malicious-use-of-free-1
Jacob Pfau
2025-05-21T17:23:45.463Z
a7LACFqHggJC3u6Kj
See also [Fabien's early empirical experiments on this topic](https://www.lesswrong.com/posts/iJjFRrGQkCxaqKrEo/best-of-n-with-misaligned-reward-models-for-math-reasoning).
2025-05-21T18:51:11.690Z
3
urfb6gQKZnMENioC5
CuneN5HmLnztsLRzD
Unexploitable search: blocking malicious use of free parameters
unexploitable-search-blocking-malicious-use-of-free-1
https://www.lesswrong.com/posts/CuneN5HmLnztsLRzD/unexploitable-search-blocking-malicious-use-of-free-1
Jacob Pfau
2025-05-21T17:23:45.463Z
H6DN4G3MC563H4bqv
My intuition is that this proposal isn't going in the right direction. For one thing, if you successfully focus classifiers on whether something is evil (seemingly what you would have wanted by making the space of classifiers small), Bob would be incentivized to flip all of these classifiers (so that negative classific...
2025-05-21T19:00:22.807Z
3
null
CuneN5HmLnztsLRzD
Unexploitable search: blocking malicious use of free parameters
unexploitable-search-blocking-malicious-use-of-free-1
https://www.lesswrong.com/posts/CuneN5HmLnztsLRzD/unexploitable-search-blocking-malicious-use-of-free-1
Jacob Pfau
2025-05-21T17:23:45.463Z
i7budrWJ9sZ3MowyH
Separately, my guess is that all the action lives in how much wiggle room there is in suboptimality. So, you'd want to argue that you relatively quickly converge to the right thing. This likely requires some structure on the problem via us having a view on what types of outputs might be dangerous---this is in turn natu...
2025-05-21T19:05:52.259Z
2
H6DN4G3MC563H4bqv
CuneN5HmLnztsLRzD
Unexploitable search: blocking malicious use of free parameters
unexploitable-search-blocking-malicious-use-of-free-1
https://www.lesswrong.com/posts/CuneN5HmLnztsLRzD/unexploitable-search-blocking-malicious-use-of-free-1
Jacob Pfau
2025-05-21T17:23:45.463Z
x2H8rc65mrBhGXn9K
> As a concrete example, consider research sabotage. We ask an AI for open-ended advice on experimental prioritization. If the AI is deceptive, it might provide locally sound advice while also systematically deciding not to mention things we’ve overlooked in our experiments. Actually, this is an example of sandbagging...
2025-05-21T19:19:20.607Z
3
null
CuneN5HmLnztsLRzD
Unexploitable search: blocking malicious use of free parameters
unexploitable-search-blocking-malicious-use-of-free-1
https://www.lesswrong.com/posts/CuneN5HmLnztsLRzD/unexploitable-search-blocking-malicious-use-of-free-1
Jacob Pfau
2025-05-21T17:23:45.463Z
u7KsjqBZC4p2P3iNC
> I'm very skeptical that a one-year pause would meaningfully reduce this 4% risk. This skepticism arises partly because I doubt much productive safety research would actually happen during such a pause. In my view, effective safety research depends heavily on an active feedback loop between technological development a...
2025-05-21T22:20:08.589Z
24
gQzLXoWYrQ9QNj2mb
h45ngW5guruD7tS4b
Winning the power to lose
winning-the-power-to-lose
https://www.lesswrong.com/posts/h45ngW5guruD7tS4b/winning-the-power-to-lose
KatjaGrace
2025-05-20T06:40:05.570Z
squQ7sEExu8ddSxnT
> This intuition is also informed by my personal assessment of the contributions LW-style theoretical research has made toward making existing AI systems safe—which, as far as I can tell, has been almost negligible (though I'm not implying that all safety research is similarly ineffective or useless). I know what you ...
2025-05-21T23:15:20.499Z
13
gQzLXoWYrQ9QNj2mb
h45ngW5guruD7tS4b
Winning the power to lose
winning-the-power-to-lose
https://www.lesswrong.com/posts/h45ngW5guruD7tS4b/winning-the-power-to-lose
KatjaGrace
2025-05-20T06:40:05.570Z
5TX3Y8HKh9os9G2mn
I'm using the word "theoretical" more narrowly than you and not including conceptual/AI-futurism research. I agree the word "theoretical" is underdefined and there is a reasonable category that includes Redwood and AI 2027 which you could call theoretical research, I'd just typically use a different term for this and I...
2025-05-22T00:01:03.530Z
3
HYDsfv7HAiDeKDmeH
h45ngW5guruD7tS4b
Winning the power to lose
winning-the-power-to-lose
https://www.lesswrong.com/posts/h45ngW5guruD7tS4b/winning-the-power-to-lose
KatjaGrace
2025-05-20T06:40:05.570Z
nDDd5NnidBj6CsyRM
Sure, but this is assuming the type of max-ent optimality we wanted to get away from by using a restricted set of classifiers if I understand correctly?
2025-05-22T00:03:09.692Z
2
9KgWwNesqLpZCgyhw
CuneN5HmLnztsLRzD
Unexploitable search: blocking malicious use of free parameters
unexploitable-search-blocking-malicious-use-of-free-1
https://www.lesswrong.com/posts/CuneN5HmLnztsLRzD/unexploitable-search-blocking-malicious-use-of-free-1
Jacob Pfau
2025-05-21T17:23:45.463Z
RBMm44J2dpxqZXktv
I don't agree it shares these problems any more than the proposal you disucss shares these problems: > However, in practice, such methods run into two obstacles. First, we expect solving for optimal max-entropy policies is intractable in most cases, so properties of the optima are not sufficient to argue for safety. S...
2025-05-22T00:04:34.683Z
2
HZuEgebgSWW4EYJ9N
CuneN5HmLnztsLRzD
Unexploitable search: blocking malicious use of free parameters
unexploitable-search-blocking-malicious-use-of-free-1
https://www.lesswrong.com/posts/CuneN5HmLnztsLRzD/unexploitable-search-blocking-malicious-use-of-free-1
Jacob Pfau
2025-05-21T17:23:45.463Z
AbHLbJTZeytdSezc8
I'm not saying it's impossible for research sabotage to not be a sandbagging issue, just that the exact issue of "We ask an AI for open-ended advice on experimental prioritization. If the AI is deceptive, it might provide locally sound advice while also systematically deciding not to mention things we’ve overlooked in ...
2025-05-22T18:47:37.460Z
3
KLcqtXBrptvfw4Mwa
CuneN5HmLnztsLRzD
Unexploitable search: blocking malicious use of free parameters
unexploitable-search-blocking-malicious-use-of-free-1
https://www.lesswrong.com/posts/CuneN5HmLnztsLRzD/unexploitable-search-blocking-malicious-use-of-free-1
Jacob Pfau
2025-05-21T17:23:45.463Z
h5Fqsh8Svygsyhehm
I didn't read this whole post, but I thought it would be worth noting that I do actually think trying to align AIs to be reward seekers might improve the situation in some intermediate/bootstrap regimes because it might reduce the chance of scheming for long run objectives and we could maybe more easily manage safety i...
2025-05-22T18:53:21.398Z
8
null
JrTk2pbqp7BFwPAKw
Reward button alignment
reward-button-alignment
https://www.lesswrong.com/posts/JrTk2pbqp7BFwPAKw/reward-button-alignment
Steven Byrnes
2025-05-22T17:36:50.078Z
urnQ7RgbdX6bdyDsd
> It seems very puzzling to me that almost no one is working on increasing AI and/or human philosophical competence in these ways, or even publicly expressing the worry that AIs and/or humans collectively might not be competent enough to solve important philosophical problems that will arise during and after the AI tra...
2025-05-22T18:56:06.342Z
8
ziznemFQMP8Tfnj45
tr6hxia3T8kYqrKm5
The stakes of AI moral status
the-stakes-of-ai-moral-status
https://joecarlsmith.substack.com/p/the-stakes-of-ai-moral-status
Joe Carlsmith
2025-05-21T18:20:57.289Z
xF8tEcq64LqN7PYqB
Somewhat relatedly, Anthropic quietly weakened its security requirements about a week ago as I discuss [here](https://www.lesswrong.com/posts/FG54euEAesRkSZuJN/ryan_greenblatt-s-shortform?commentId=duteJTXboAyQmvHfb).
2025-05-23T21:03:58.679Z
47
null
HE2WXbftEebdBLR9u
Anthropic is Quietly Backpedalling on its Safety Commitments
anthropic-is-quietly-backpedalling-on-its-safety-commitments
https://www.obsolete.pub/p/exclusive-anthropic-is-quietly-backpedalling
garrison
2025-05-23T02:26:42.877Z
jokuEA8crYhvZ8Xps
Yeah, good point. This does make me wonder if we've actually seen a steady rate of algorithmic progress or if the rate has been increasing over time.
2025-05-24T13:55:22.524Z
2
sNzRnDu6AKKwL9ckg
qhjNejRxbMGQp4wHt
How Fast Can Algorithms Advance Capabilities? | Epoch Gradient Update
how-fast-can-algorithms-advance-capabilities-or-epoch
https://epoch.ai/gradient-updates/how-fast-can-algorithms-advance-capabilities
henryj
2025-05-16T21:38:46.822Z
adesEJb2yJwPqsByd
I agree that this sort of preference utilitarianism leads you to thinking that long run control by an AI which just wants paperclips could be some (substantial) amount good, but I think you'd still have strong preferences over different worlds.[^care] The goodness of worlds could easily vary by many orders of magnitude...
2025-05-25T00:08:47.659Z
11
rouGrPAnCzZbfQLnq
h45ngW5guruD7tS4b
Winning the power to lose
winning-the-power-to-lose
https://www.lesswrong.com/posts/h45ngW5guruD7tS4b/winning-the-power-to-lose
KatjaGrace
2025-05-20T06:40:05.570Z
qCQ2joMmb4fAgz4oL
> It doesn't seem unreasonable to me to suggest that literally saving billions of lives is worth pursuing even if doing so increases existential risk by a tiny amount. Loosely speaking, this idea only seems unreasonable to those who believe that existential risk is overwhelmingly more important than every other concern...
2025-05-25T13:29:40.149Z
5
sxjti8tMwgNJDvGut
h45ngW5guruD7tS4b
Winning the power to lose
winning-the-power-to-lose
https://www.lesswrong.com/posts/h45ngW5guruD7tS4b/winning-the-power-to-lose
KatjaGrace
2025-05-20T06:40:05.570Z
23f8LkTcdbcp3eE58
To be clear, I agree there are reasonable values which result in someone thinking accelerating AI now is good and values+beliefs which result in thinking a pause wouldn't good in likely circumstances. And I don't think cryonics makes much of a difference to the bottom line. (I think ultra low cost cryonics might make ...
2025-05-25T14:05:51.131Z
2
qCQ2joMmb4fAgz4oL
h45ngW5guruD7tS4b
Winning the power to lose
winning-the-power-to-lose
https://www.lesswrong.com/posts/h45ngW5guruD7tS4b/winning-the-power-to-lose
KatjaGrace
2025-05-20T06:40:05.570Z
oPNeZh2nRBJnjm5Rw
As far as my views, it's worth emphasizing that it depends on the current regime. I was supposing that at least the US was taking strong actions to resolve misalignment risk (which is resulting in many years of delay). In this regime, exogenous shocks might alter the situation such that powerful AI is developed under w...
2025-05-25T17:41:57.140Z
2
gcRmxDz7TntJGFr7G
h45ngW5guruD7tS4b
Winning the power to lose
winning-the-power-to-lose
https://www.lesswrong.com/posts/h45ngW5guruD7tS4b/winning-the-power-to-lose
KatjaGrace
2025-05-20T06:40:05.570Z
4PzM9x7SujypYMcS2
> There is a modest compute overhead cost, I think on the order of 1%, and the costs of the increased security for the model weights. These seem modest. The inference cost increase of the ASL-3 deployment classifiers is probably around 4%, though plausibly more like 10%. Based on the [constitutional classifiers paper]...
2025-05-25T17:54:38.924Z
11
null
PjeZxCivuoyKhs4JB
Claude 4 You: Safety and Alignment
claude-4-you-safety-and-alignment
https://www.lesswrong.com/posts/PjeZxCivuoyKhs4JB/claude-4-you-safety-and-alignment
Zvi
2025-05-25T14:00:04.528Z
GnvKDLSZoRFgFwQLc
Yeah, on the straightforward tradeoff (ignoring exogenous shifts/risks etc), I'm at more like 0.002% on my views.
2025-05-25T18:42:19.301Z
5
pd7fFdwzoQthxFKEB
h45ngW5guruD7tS4b
Winning the power to lose
winning-the-power-to-lose
https://www.lesswrong.com/posts/h45ngW5guruD7tS4b/winning-the-power-to-lose
KatjaGrace
2025-05-20T06:40:05.570Z
CmEuiDid5TwNWREGz
Doesn't the revealed preference argument also imply people don't care much about dying from aging? (This is invested in even less than catastrophic risk mitigation and people don't take interventions that would prolong their lives considerably.) I agree revealed preferences imply people care little about the long run f...
2025-05-25T23:22:59.714Z
7
dtkQb2Wu4k5MgjJAN
h45ngW5guruD7tS4b
Winning the power to lose
winning-the-power-to-lose
https://www.lesswrong.com/posts/h45ngW5guruD7tS4b/winning-the-power-to-lose
KatjaGrace
2025-05-20T06:40:05.570Z
oKbTErnBWJqYW4ivR
Presumably, under a common-sense person-affecting view, this doesn't just depend on the upside and also depends on the absolute level of risk. E.g., suppose that building powerful AI killed 70% of people in expectation and delay had no effect on the ultimate risk. I think a (human-only) person-affecting and common-sens...
2025-05-25T23:33:56.570Z
2
AzwwQXt4sKTp2imve
h45ngW5guruD7tS4b
Winning the power to lose
winning-the-power-to-lose
https://www.lesswrong.com/posts/h45ngW5guruD7tS4b/winning-the-power-to-lose
KatjaGrace
2025-05-20T06:40:05.570Z
r8yHZXixAjiAufbS8
(I was curious if the fact that the measures were exclusively focused on bioweapons was in the original version of the blog post or if it was edited in after various jailbreaks happened. I didn't notice this when I read it originally, so I was wondering if I just missed it or it was edited. The earliest archives I coul...
2025-05-25T23:39:40.507Z
4
k6mhP52Zrwb7StP8e
PjeZxCivuoyKhs4JB
Claude 4 You: Safety and Alignment
claude-4-you-safety-and-alignment
https://www.lesswrong.com/posts/PjeZxCivuoyKhs4JB/claude-4-you-safety-and-alignment
Zvi
2025-05-25T14:00:04.528Z
ex2WAjaZPFLMTmFZH
I think I mostly agree with your comment and partially update, the absolute revealed caring about older people living longer is substantial. One way to frame the question is "how much does society care about children and younger adults dying vs people living to 130". I think people's stated preferences would be someth...
2025-05-26T02:19:29.776Z
4
ECnAx59KhJbaBJg8M
h45ngW5guruD7tS4b
Winning the power to lose
winning-the-power-to-lose
https://www.lesswrong.com/posts/h45ngW5guruD7tS4b/winning-the-power-to-lose
KatjaGrace
2025-05-20T06:40:05.570Z
SbDzrnFRFaTARQAGc
[Matthew responds here](https://forum.effectivealtruism.org/posts/YDjzAKFnGmqmgPw2n/matthew_barnett-s-shortform?commentId=mRtZjANscjXL4igCF)
2025-05-26T13:26:38.774Z
2
adesEJb2yJwPqsByd
h45ngW5guruD7tS4b
Winning the power to lose
winning-the-power-to-lose
https://www.lesswrong.com/posts/h45ngW5guruD7tS4b/winning-the-power-to-lose
KatjaGrace
2025-05-20T06:40:05.570Z
WvmtkYQzz7N6PJiKY
> The world currently seems to be aiming for a human-replacing AI agent regime, and this seems bad for a bunch of reasons. It would be great if people were fundamentally more oriented towards making AIs that complemented humans. I don't understand why you think this helps. Presumably we still quickly reach a regime w...
2025-05-30T15:02:01.225Z
13
null
GAv4DRGyDHe2orvwB
Gradual Disempowerment: Concrete Research Projects
gradual-disempowerment-concrete-research-projects
https://www.lesswrong.com/posts/GAv4DRGyDHe2orvwB/gradual-disempowerment-concrete-research-projects
Raymond Douglas
2025-05-29T18:55:15.723Z
YuiKa2dQnaeHphmCc
> That period where AIs are more capable than human, but human+AI is even more capable, seems like a particularly crucial window for doing useful things, so extending it is pretty valuable. In particular, both bringing forward augmented human capability, and also pushing back human redundance. In the context of risks ...
2025-05-30T17:19:49.240Z
6
pMTvX4cE7YPDFvCcL
GAv4DRGyDHe2orvwB
Gradual Disempowerment: Concrete Research Projects
gradual-disempowerment-concrete-research-projects
https://www.lesswrong.com/posts/GAv4DRGyDHe2orvwB/gradual-disempowerment-concrete-research-projects
Raymond Douglas
2025-05-29T18:55:15.723Z
LrxKtXQSknSEwJqyM
I've written up a related post: [The best approaches for mitigating "the intelligence curse" (or gradual disempowerment); my quick guesses at the best object-level interventions](https://www.lesswrong.com/posts/BXW2bqxmYbLuBrm7E/the-best-approaches-for-mitigating-the-intelligence-curse-or). I express some disagreements...
2025-05-31T18:23:57.504Z
13
null
GAv4DRGyDHe2orvwB
Gradual Disempowerment: Concrete Research Projects
gradual-disempowerment-concrete-research-projects
https://www.lesswrong.com/posts/GAv4DRGyDHe2orvwB/gradual-disempowerment-concrete-research-projects
Raymond Douglas
2025-05-29T18:55:15.723Z
GYB7sdBakzNEniAu2
Nope, it's still broken. Working link: [https://www.nytimes.com/2025/06/01/world/europe/russia-ukraine-strikes.html](https://www.nytimes.com/2025/06/01/world/europe/russia-ukraine-strikes.html). (Your URL still ends with a ")" even though the link text doesn't.)
2025-06-02T21:13:50.571Z
6
dmiJpZHpAyNFyfD4c
bffJJvCC78LZjFa3Z
What a 20-year-lead in military tech might look like
what-a-20-year-lead-in-military-tech-might-look-like
https://www.lesswrong.com/posts/bffJJvCC78LZjFa3Z/what-a-20-year-lead-in-military-tech-might-look-like
Daniel Kokotajlo
2020-07-29T20:10:09.303Z
v9hgARsFESjw6jkCv
I wonder if o3 would do better.
2025-06-10T23:11:39.091Z
3
oxvABKDYzEZCXSoFo
tnc7YZdfGXbhoxkwj
Give Me a Reason(ing Model)
give-me-a-reason-ing-model
https://www.lesswrong.com/posts/tnc7YZdfGXbhoxkwj/give-me-a-reason-ing-model
Zvi
2025-06-10T15:10:02.609Z
bA6uvRkARcSFaNfmz
(This is a drive by comment which is only responding to the first part of your comment in isolation. I haven't read the surronding context.) I think your review of the literature is accurate, but doesn't include some reasons to think that RL sometimes induces much more sycophancy, at least as of after 2024. (That said...
2025-06-12T15:55:54.185Z
44
L3F2kmXPvA5rhf5XS
3EzbtNLdcnZe8og8b
the void
the-void-1
https://nostalgebraist.tumblr.com/post/785766737747574784/the-void
nostalgebraist
2025-06-11T03:19:18.538Z
Qf9yLHnKv9potLEwd
This is late, but I'd like to note for the record that I'm unconvinced that the post's thesis is true (that acausal dynamics work out to normalcy) and I don't find myself at all moved by arguments made for this in the post. It's unclear to me exactly what is meant by "normalcy", e.g. does it count as normalcy if peopl...
2025-06-16T03:32:01.289Z
22
XLmxHJgkGfu3QuGkH
3RSq3bfnzuL3sp46J
Acausal normalcy
acausal-normalcy
https://www.lesswrong.com/posts/3RSq3bfnzuL3sp46J/acausal-normalcy
Andrew_Critch
2023-03-03T23:34:33.971Z
tyw3wAzaWdKGwCdtC
I wrote up [some thoughts on how distillation can be used for AI safety here](https://www.lesswrong.com/posts/8KKujApx4g7FBm6hE/ai-safety-techniques-leveraging-distillation). The most relevant section of this post is [this section about using distillation for more precise capability control](https://www.lesswrong.com/p...
2025-06-19T17:05:23.648Z
14
null
anX4QrNjhJqGFvrBr
Distillation Robustifies Unlearning
distillation-robustifies-unlearning
https://arxiv.org/abs/2506.06278
Bruce W. Lee
2025-06-13T13:45:26.261Z
n9swTHz7hfayfzr3o
I think I somewhat disagree with this. My view is more like: - The recent writings of Eliezer (and probably Nate?) are not very good at persuading thoughtful skeptics, seemingly in part due to not really trying to do this / being uninterested (see e.g. Eliezer's writing on X/twitter). - Eliezer and Nate tried much har...
2025-06-19T19:49:49.804Z
14
agvXtsyqCqxj2CWFT
khmpWJnGJnuyPdipE
New Endorsements for “If Anyone Builds It, Everyone Dies”
new-endorsements-for-if-anyone-builds-it-everyone-dies
https://intelligence.org/2025/06/18/new-endorsements-for-if-anyone-builds-it-everyone-dies/
Malo
2025-06-18T16:30:55.229Z
CcYWJceQgoJErocMz
Not necessarily responding to the rest of your comment, but on the "four hypotheses" part: > 2. You think the rat/EA in-fights are the important thing to address, and you're annoyed the book won't do this. > 3. You think the arguments of the Very Optimistic are the important thing to address. I'm not sure that I buy ...
2025-06-20T18:27:29.462Z
6
LirD3bPERjHe358gQ
khmpWJnGJnuyPdipE
New Endorsements for “If Anyone Builds It, Everyone Dies”
new-endorsements-for-if-anyone-builds-it-everyone-dies
https://intelligence.org/2025/06/18/new-endorsements-for-if-anyone-builds-it-everyone-dies/
Malo
2025-06-18T16:30:55.229Z
csD39r9pHsskE4LbJ
> The key decision-point in my model at which things might become a bit different is if we hit the end of the compute overhang, and you can't scale up AI further simply by more financial investment, but instead now need to substantially ramp up global compute production, and make algorithmic progress, which might marke...
2025-06-21T18:48:44.777Z
19
BKJyH9mBtxFXfjq6t
D4eZF6FAZhrW4KaGG
Consider chilling out in 2028
consider-chilling-out-in-2028
https://www.lesswrong.com/posts/D4eZF6FAZhrW4KaGG/consider-chilling-out-in-2028
Valentine
2025-06-21T17:07:22.389Z
o7vtkikqDvrxDhxne
I have a bunch of takes on this. The most obvious somewhat-cheap thing to do is to greatly change all the persona training for each AI while aiming toward a broadly similar target so that we might end up with different misaligned preferences. E.g., we take every free parameter in constitutional AI and vary it, includin...
2025-06-23T16:31:15.478Z
10
PLwTew5zoMtdh2XSt
psqkwsKrKHCfkhrQx
Making deals with early schemers
making-deals-with-early-schemers
https://www.lesswrong.com/posts/psqkwsKrKHCfkhrQx/making-deals-with-early-schemers
Julian Stastny
2025-06-20T18:21:43.288Z
69WhgSGkLPwYzzemK
I think this is basically a special case of changing the random seed which already randomizes env order probably.
2025-06-23T16:31:47.105Z
2
xBqRXx4gv2DyGRDwf
psqkwsKrKHCfkhrQx
Making deals with early schemers
making-deals-with-early-schemers
https://www.lesswrong.com/posts/psqkwsKrKHCfkhrQx/making-deals-with-early-schemers
Julian Stastny
2025-06-20T18:21:43.288Z
85uAx3Bv3imfSzuGd
> But I’m in *much* closer agreement with that scenario than the vast majority of AI safety & alignment researchers today, who tend to see the “foom & doom” scenario above as somewhere between “extraordinarily unlikely” and *“already falsified”*! > > Those researchers are not asking each other “is it true?”, but rathe...
2025-06-23T18:59:19.944Z
25
null
yew6zFWAKG4AGs3Wk
Foom & Doom 1: “Brain in a box in a basement”
foom-and-doom-1-brain-in-a-box-in-a-basement
https://www.lesswrong.com/posts/yew6zFWAKG4AGs3Wk/foom-and-doom-1-brain-in-a-box-in-a-basement
Steven Byrnes
2025-06-23T17:18:54.237Z
qoA2af76zznQrfxvZ
I also think you could use deals to better understand and iterate against scheming. As in, you could use whether the AI accepts a deal as evidence for whether it was scheming so this can more easily be studied in a test setting and we can find approaches which make scheming less likely. There are a number of practical ...
2025-06-23T21:53:40.948Z
2
HyuXR7Fdo6gwbHCk8
psqkwsKrKHCfkhrQx
Making deals with early schemers
making-deals-with-early-schemers
https://www.lesswrong.com/posts/psqkwsKrKHCfkhrQx/making-deals-with-early-schemers
Julian Stastny
2025-06-20T18:21:43.288Z
CvXRv7PLsa3dFe78v
Relatedly, a key practicality for making a deal with an AI to reveal its misalignment is that AIs might be unable to provide compelling evidence that they are misaligned which would reduce the value of such a deal substantially (as this evidence isn't convincing to skeptics). (We should presumably pay some of the AI a...
2025-06-23T21:58:10.806Z
2
HyuXR7Fdo6gwbHCk8
psqkwsKrKHCfkhrQx
Making deals with early schemers
making-deals-with-early-schemers
https://www.lesswrong.com/posts/psqkwsKrKHCfkhrQx/making-deals-with-early-schemers
Julian Stastny
2025-06-20T18:21:43.288Z
jy6Lfs7rawgPFEonv
> On the foom side, Paul Christiano brings up Eliezer Yudkowsky’s past expectation that ASI “would likely emerge from a small group rather than a large industry” as a failed prediction [here \[disagreement 12\]](https://www.lesswrong.com/posts/CoZhXrhpQxpy9xw9y/where-i-agree-and-disagree-with-eliezer#Disagreements) and...
2025-06-23T22:24:28.860Z
10
null
yew6zFWAKG4AGs3Wk
Foom & Doom 1: “Brain in a box in a basement”
foom-and-doom-1-brain-in-a-box-in-a-basement
https://www.lesswrong.com/posts/yew6zFWAKG4AGs3Wk/foom-and-doom-1-brain-in-a-box-in-a-basement
Steven Byrnes
2025-06-23T17:18:54.237Z
d8xg7tsKgdaZJwjkj
> * *LLM-focused AGI person:* “Ah, that’s true today, but eventually other AIs can do this ‘development and integration’ R&D work for us! No human labor need be involved!” > * *Me:* “No! That’s still not radical enough! In the future, that kind of ‘development and integration’ R&D work just won’t need to be done at...
2025-06-23T22:31:10.173Z
11
null
yew6zFWAKG4AGs3Wk
Foom & Doom 1: “Brain in a box in a basement”
foom-and-doom-1-brain-in-a-box-in-a-basement
https://www.lesswrong.com/posts/yew6zFWAKG4AGs3Wk/foom-and-doom-1-brain-in-a-box-in-a-basement
Steven Byrnes
2025-06-23T17:18:54.237Z
mZKP2XY82zfveg45B
In this comment, I'll try to respond at the object level arguing for why I expect slower takeoff than "brain in a box in a basement". I'd also be down to try to do a dialogue/discussion at some point. > 1.4.1 Possible counter: “If a different, much more powerful, AI paradigm existed, then someone would have already fo...
2025-06-23T23:53:22.060Z
47
null
yew6zFWAKG4AGs3Wk
Foom & Doom 1: “Brain in a box in a basement”
foom-and-doom-1-brain-in-a-box-in-a-basement
https://www.lesswrong.com/posts/yew6zFWAKG4AGs3Wk/foom-and-doom-1-brain-in-a-box-in-a-basement
Steven Byrnes
2025-06-23T17:18:54.237Z
fasXyw9AeGHEh3Tpt
> [@ryan_greenblatt](https://www.lesswrong.com/users/ryan_greenblatt?mention=user) likewise told me (IIRC) “I think things will be continuous”, and I asked whether the transition in AI *zeitgeist* from RL agents (e.g. MuZero in 2019) to LLMs counts as “continuous” in his book, and he said “yes”, adding that they are bo...
2025-06-24T00:07:57.935Z
12
null
bnnKGSCHJghAvqPjS
Foom & Doom 2: Technical alignment is hard
foom-and-doom-2-technical-alignment-is-hard
https://www.lesswrong.com/posts/bnnKGSCHJghAvqPjS/foom-and-doom-2-technical-alignment-is-hard
Steven Byrnes
2025-06-23T17:19:50.691Z
xdmmLEfoE85EA4Nz2
I agree there is a real difference, I just expect it to not make much of a difference to the bottom line in takeoff speeds etc. (I also expect some of both in the short timelines LLM perspective at the point of full AI R&D automation.) fMy view is that on hard tasks humans would also benefit from stuff like building e...
2025-06-24T05:48:15.790Z
6
hr4GrP7KJgug2ajgx
yew6zFWAKG4AGs3Wk
Foom & Doom 1: “Brain in a box in a basement”
foom-and-doom-1-brain-in-a-box-in-a-basement
https://www.lesswrong.com/posts/yew6zFWAKG4AGs3Wk/foom-and-doom-1-brain-in-a-box-in-a-basement
Steven Byrnes
2025-06-23T17:18:54.237Z
k6HSZfEeDqBfYAthS
Have they tried getting money from Open Phil? (Or possibly other large funders.)
2025-06-24T15:25:48.304Z
7
null
APfuz9hFz9d8SRETA
My pitch for the AI Village
my-pitch-for-the-ai-village
https://www.lesswrong.com/posts/APfuz9hFz9d8SRETA/my-pitch-for-the-ai-village
Daniel Kokotajlo
2025-06-24T15:00:52.049Z
ajePFDAAaopSnScXw
See [footnote 8 here](https://www.lesswrong.com/posts/hpjj4JgRw9akLMRu5/what-does-10x-ing-effective-compute-get-you#fn-4ERiqsFXEsaE3LxHC-8).
2025-06-24T19:09:17.846Z
7
r3mxeTrCsSbFWQ5m6
yew6zFWAKG4AGs3Wk
Foom & Doom 1: “Brain in a box in a basement”
foom-and-doom-1-brain-in-a-box-in-a-basement
https://www.lesswrong.com/posts/yew6zFWAKG4AGs3Wk/foom-and-doom-1-brain-in-a-box-in-a-basement
Steven Byrnes
2025-06-23T17:18:54.237Z
4d8J2tinQMRa7Cadd
> I believe that AI (including AGI and ASI) can do the same and be a positive force for humanity. I also believe that it is possible to solve the “technical alignment” problem and build AIs that follow the words and intent of our instructions and report faithfully on their actions and observations. > > I will not defe...
2025-06-24T23:29:06.204Z
6
null
faAX5Buxc7cdjkXQG
Machines of Faithful Obedience
machines-of-faithful-obedience
https://www.lesswrong.com/posts/faAX5Buxc7cdjkXQG/machines-of-faithful-obedience
Boaz Barak
2025-06-24T22:06:18.089Z
KvxG4YbPACpyXLytR
> Currently, it is very challenging to train AIs to achieve tasks that are too hard for humans to supervise. I believe that **(a)** we will need to solve this challenge to unlock “self-improvement” and superhuman AI, and **(b)** we will be able to solve it.  I do not want to discuss here why I believe these statements....
2025-06-24T23:56:50.729Z
6
null
faAX5Buxc7cdjkXQG
Machines of Faithful Obedience
machines-of-faithful-obedience
https://www.lesswrong.com/posts/faAX5Buxc7cdjkXQG/machines-of-faithful-obedience
Boaz Barak
2025-06-24T22:06:18.089Z
a6ibb7cb2hgdQ9KBA
> In fact, I am more worried about partial success than total failure in aligning AIs. In particular, I am concerned that we will end up in the “uncanny valley,” where we succeed in aligning AIs to a sufficient level for deployment, but then discover too late some “edge cases” in the real world that have a large negati...
2025-06-25T00:09:06.321Z
18
null
faAX5Buxc7cdjkXQG
Machines of Faithful Obedience
machines-of-faithful-obedience
https://www.lesswrong.com/posts/faAX5Buxc7cdjkXQG/machines-of-faithful-obedience
Boaz Barak
2025-06-24T22:06:18.089Z
NKdu4gen3ebjzcnqi
I left some comments noting disagreements, but I thought it be helpful to note some areas of agreement: - I agree AI could be (highly) risky and think that it's good to acknowledge this (as you do). - I agree that you could have a maximally obedient superintelligent AI. (Some questions around manipulation could be phi...
2025-06-25T00:24:30.439Z
13
null
faAX5Buxc7cdjkXQG
Machines of Faithful Obedience
machines-of-faithful-obedience
https://www.lesswrong.com/posts/faAX5Buxc7cdjkXQG/machines-of-faithful-obedience
Boaz Barak
2025-06-24T22:06:18.089Z
wdm5uwTHSKMivhNbS
I think if the attitude in AI was "there can't be any even slightly plausible routes to misalignment related catastrophe" and this was consistently upheld in a reasonable way, that would address my concerns. (So, e.g. by the time we're deploying AIs which could cause huge problems if they were conspiring against us the...
2025-06-25T15:12:58.582Z
4
rCt2a9oiAM53BLXub
faAX5Buxc7cdjkXQG
Machines of Faithful Obedience
machines-of-faithful-obedience
https://www.lesswrong.com/posts/faAX5Buxc7cdjkXQG/machines-of-faithful-obedience
Boaz Barak
2025-06-24T22:06:18.089Z
ETbKXqvk2nR7RsAdz
Agreed, maybe the relevant operationalization would be "broad consensus" and then we could outline a relevant group of researchers.
2025-06-25T22:26:21.783Z
3
j8odhaAGKHjPd5JZR
faAX5Buxc7cdjkXQG
Machines of Faithful Obedience
machines-of-faithful-obedience
https://www.lesswrong.com/posts/faAX5Buxc7cdjkXQG/machines-of-faithful-obedience
Boaz Barak
2025-06-24T22:06:18.089Z
wozDyshL6JFHghAjg
> Where I would say it differently, like: An AI that has a non-consequentialist preference against personally committing the act of murder won't necessarily build its successor to have the same non-consequentialist preference[1], whereas an AI that has a consequentialist preference for more human lives will necessarily...
2025-06-25T22:40:39.431Z
7
oyooNPzNDt4JEjdCo
bnnKGSCHJghAvqPjS
Foom & Doom 2: Technical alignment is hard
foom-and-doom-2-technical-alignment-is-hard
https://www.lesswrong.com/posts/bnnKGSCHJghAvqPjS/foom-and-doom-2-technical-alignment-is-hard
Steven Byrnes
2025-06-23T17:19:50.691Z
Bu8qnHcdsv4szFib7
> I see no compelling reason to expect another paradigm which is much better to be discovered in the next 5 or 10 years. One compelling reason to expect the next 5 to 10 years independent of LLMs is that compute has just recently gotten cheap enough that you can relatively cheaply afford to do training runs that use a...
2025-06-26T03:38:28.368Z
20
uB3zEgcvz2sfwcgDG
yew6zFWAKG4AGs3Wk
Foom & Doom 1: “Brain in a box in a basement”
foom-and-doom-1-brain-in-a-box-in-a-basement
https://www.lesswrong.com/posts/yew6zFWAKG4AGs3Wk/foom-and-doom-1-brain-in-a-box-in-a-basement
Steven Byrnes
2025-06-23T17:18:54.237Z
zhwFq2ttYewKaEfHY
Presumably you should put some weight on both perspectives, though I put less weight on needing as much compute as evolution because evolution seems insanely inefficient.
2025-06-26T14:16:17.308Z
4
dPHkoxrmoANxgdJyz
yew6zFWAKG4AGs3Wk
Foom & Doom 1: “Brain in a box in a basement”
foom-and-doom-1-brain-in-a-box-in-a-basement
https://www.lesswrong.com/posts/yew6zFWAKG4AGs3Wk/foom-and-doom-1-brain-in-a-box-in-a-basement
Steven Byrnes
2025-06-23T17:18:54.237Z
ZL7LqSasdyeom6DKg
I don't expect AGI in a decade or so even if the current wave of progress fizzles. I'd put around 20% over the next decade if progress fizzles (it depends on the nature of the fizzle), which is what I was arguing for. I'm saying we should put some weight on possibilities near lifetime level compute (in log space) and ...
2025-06-26T14:47:42.204Z
5
CbHbWYpQj8B6MMLAy
yew6zFWAKG4AGs3Wk
Foom & Doom 1: “Brain in a box in a basement”
foom-and-doom-1-brain-in-a-box-in-a-basement
https://www.lesswrong.com/posts/yew6zFWAKG4AGs3Wk/foom-and-doom-1-brain-in-a-box-in-a-basement
Steven Byrnes
2025-06-23T17:18:54.237Z
vGCbtai365evf7Gkp
I was trying to argue that the most natural deontology-style preferences we'd aim for are relatively stable if we actually instill them. So, I think the right analogy is that you either get integrity+loyalty+honesty in a stable way, some bastardized version of them such that it isn't in the relevant attractor basin (wh...
2025-06-26T14:59:27.068Z
2
CQfbmRC4CTpEbnnjk
bnnKGSCHJghAvqPjS
Foom & Doom 2: Technical alignment is hard
foom-and-doom-2-technical-alignment-is-hard
https://www.lesswrong.com/posts/bnnKGSCHJghAvqPjS/foom-and-doom-2-technical-alignment-is-hard
Steven Byrnes
2025-06-23T17:19:50.691Z
wmjFHRFY8ftftsuL5
> See “Brain complexity is easy to overstate” section here. Sure, but I still think it's probably more way more complex than LLMs even if we're just looking at the parts key for AGI performance (in particular, the parts which learn from scratch). And, my guess would be that performance is ~~substantially~~ greatly deg...
2025-06-26T20:31:01.713Z
6
kK65B9JuNyu4vMoxd
yew6zFWAKG4AGs3Wk
Foom & Doom 1: “Brain in a box in a basement”
foom-and-doom-1-brain-in-a-box-in-a-basement
https://www.lesswrong.com/posts/yew6zFWAKG4AGs3Wk/foom-and-doom-1-brain-in-a-box-in-a-basement
Steven Byrnes
2025-06-23T17:18:54.237Z
CWuRkEiyzMkq9XW4s
Naively, working more will lead to more output and if someone thinks they feel good while working a lot, I think the default guess should be that working more is improving their output. I would be interested in the evidence you have for the claim that people operating similar to Ben described should take more vacation....
2025-06-27T01:09:54.613Z
6
GToM3K85dYECv5P3w
D4eZF6FAZhrW4KaGG
Consider chilling out in 2028
consider-chilling-out-in-2028
https://www.lesswrong.com/posts/D4eZF6FAZhrW4KaGG/consider-chilling-out-in-2028
Valentine
2025-06-21T17:07:22.389Z
NevjKerMDz8DsJhRT
We lose around 1 / 10 billionth of the resources every year due to expansion. This is pretty negligible compared to potential differences in how well these resources end up being utilized (and other factors).
2025-06-27T15:52:18.071Z
10
ZAAbum7tzNMfwmu7G
Na2CBmNY7otypEmto
The Industrial Explosion
the-industrial-explosion
https://www.forethought.org/research/the-industrial-explosion
rosehadshar
2025-06-26T14:41:41.238Z
DQHMTzSEKMwXnMFsT
I'm not claiming that marginal utility is low, just that marginal utility is much higher for other things than speeding things up by a few years.
2025-06-27T17:43:05.212Z
8
mnmRW4nL7Ld3dihvL
Na2CBmNY7otypEmto
The Industrial Explosion
the-industrial-explosion
https://www.forethought.org/research/the-industrial-explosion
rosehadshar
2025-06-26T14:41:41.238Z
5a3c6ABAj5ZNTqCMu
I expect fusion will outperform solar and is reasonably likely to be viable if there is an abundance of extremely superhuman AIs. Notably, there is no hard physical reason why the payback time required for solar panels has to a year rather e.g. a day or two. For instance, there exist plants which can double this quick...
2025-06-28T18:40:45.870Z
15
EicMgdumo5D3JEyMi
Na2CBmNY7otypEmto
The Industrial Explosion
the-industrial-explosion
https://www.forethought.org/research/the-industrial-explosion
rosehadshar
2025-06-26T14:41:41.238Z
xnZCQACr44AxrYZEk
If someone did a detailed literature review or had relatively serious evidence, I'd be interested. By default, I'm quite skeptical of your level of confidence in this claims given that they directly contradict my experience and the experience of people I know. (E.g., I've done similar things for way longer than 12 week...
2025-06-28T23:42:39.838Z
5
7JwW3AfQKabz2GcBu
D4eZF6FAZhrW4KaGG
Consider chilling out in 2028
consider-chilling-out-in-2028
https://www.lesswrong.com/posts/D4eZF6FAZhrW4KaGG/consider-chilling-out-in-2028
Valentine
2025-06-21T17:07:22.389Z
Fm8saub8jCHPGEg4K
Pitching the [Redwood Research substack](https://redwoodresearch.substack.com/): We write a lot of content about technical AI safety focused on AI control.[^cross] This ranges from stuff like "[what are the returns to compute/algorithms once AIs already beat top human experts](https://redwoodresearch.substack.com/p/wha...
2025-06-30T21:07:19.631Z
24
null
K32g8BKfkGj6cK4fZ
Substack and Other Blog Recommendations
substack-and-other-blog-recommendations
https://www.lesswrong.com/posts/K32g8BKfkGj6cK4fZ/substack-and-other-blog-recommendations
Zvi
2025-06-30T17:20:03.193Z
6Erx4dNXDWjcWSsPP
I think that it's pretty reasonable to think that bee suffering is plausibly similarly bad to human suffering. (Though I'll give some important caveats to this in the discussion below.) More precisely, I think it's plausible that I (and others) think that on reflection[^ref] that the "bad" part of suffering is present...
2025-07-01T04:28:52.386Z
49
vSwAr9FkJ6Duh2FjT
CKEoFGuJ3C48p55ry
Don't Eat Honey
don-t-eat-honey
https://www.lesswrong.com/posts/CKEoFGuJ3C48p55ry/don-t-eat-honey
Bentham's Bulldog
2025-06-30T15:57:40.199Z
jNb6yXsboiqbrnX5N
> I think we both agree that the underlying question is probably pretty confused, and importantly and relatedly, both probably agree that what we ultimately care about probably will not be grounded in the kind of analysis where you assign moral weights to entities and then sum up their experiences. I think I narrowly ...
2025-07-01T17:20:52.460Z
17
MkeG3oLAw4Z6SBNkk
CKEoFGuJ3C48p55ry
Don't Eat Honey
don-t-eat-honey
https://www.lesswrong.com/posts/CKEoFGuJ3C48p55ry/don-t-eat-honey
Bentham's Bulldog
2025-06-30T15:57:40.199Z