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 |
|---|---|---|---|---|---|---|---|---|---|---|
7kduyL2bbxzaSChZ4 | Hmm, I guess I think "something basically like hedonic utilitarianism, at least for downside" is pretty plausible.
Maybe a big difference is that I feel like I've generally updated away from putting much weight on moral intuitions / heuristics *except with respect to forbidding some actions because they violate norms,... | 2025-07-01T19:37:39.949Z | 10 | JuXAFwckwnvhLBkSS | 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 |
zDttL73gwkp6EDqwN | oops, I meant except. My terrible spelling strikes again. | 2025-07-01T20:32:11.678Z | 2 | xmtreDZ5XvRJAxfNp | 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 |
KyHjPygMkdrKfboRa | > First blood times represent the time of first successful submission in the originally published competition. While there are some limitations (participants usually compete in teams, and may solve problems in parallel or in sequence), this still provides a useful proxy.
Another limitation is that first blood times re... | 2025-07-02T19:30:38.846Z | 4 | null | fjgYkTWKAXSxsxdsj | AI Task Length Horizons in Offensive Cybersecurity | ai-task-length-horizons-in-offensive-cybersecurity | https://www.lesswrong.com/posts/fjgYkTWKAXSxsxdsj/ai-task-length-horizons-in-offensive-cybersecurity | Sean Peters | 2025-07-02T18:36:45.126Z |
JYyKAWsQQ6Txrv6cS | > Adopting new hardware will require modifying security-critical code
> -------------------------------------------------------------------
Another concern is that AI companies (or the AI company) will rapidly buy a bunch of existing hardware (GPUs, other accelerators, etc.) during the singularity, and handling this h... | 2025-07-04T17:58:44.291Z | 6 | null | qKz2hBahahmb4uDty | How much novel security-critical infrastructure do you need during the singularity? | how-much-novel-security-critical-infrastructure-do-you-need | https://www.lesswrong.com/posts/qKz2hBahahmb4uDty/how-much-novel-security-critical-infrastructure-do-you-need | Buck | 2025-07-04T16:54:53.285Z |
ZuFYCNYAuzpWsv5va | > The bar for ‘good enough’ might be quite high
Presumably a key assumption of this strategy is that takeoff is slow enough that AIs which are good enough at improving collective epistemics and coordination are sufficiently cheap and sufficiently available before it's too late. | 2025-07-04T18:04:44.617Z | 6 | null | kvZyCJ4qMihiJpfCr | ‘AI for societal uplift’ as a path to victory | ai-for-societal-uplift-as-a-path-to-victory | https://www.lesswrong.com/posts/kvZyCJ4qMihiJpfCr/ai-for-societal-uplift-as-a-path-to-victory | Raymond Douglas | 2025-07-04T15:32:38.384Z |
ofeT8ooL3pKR2rtZD | It seems good to spend a bunch of time on takeoff speeds given how important they are for how AI goes. There are many sources discussing takeoff speeds. Some places to look: [Tom Davidson's outputs](https://www.lesswrong.com/users/tom-davidson-1), [forethought](https://www.forethought.org/), [AI-2027 (including supplem... | 2025-07-05T03:28:40.729Z | 2 | null | qe8LjXAtaZfrc8No7 | Call for suggestions - AI safety course | call-for-suggestions-ai-safety-course | https://www.lesswrong.com/posts/qe8LjXAtaZfrc8No7/call-for-suggestions-ai-safety-course | Boaz Barak | 2025-07-03T14:30:41.147Z |
2guPiC4S8qeKEFXiz | I agree that my view is that they can count as continuous (though the exact definition of the word continuous can matter!), but then the statement "I find this perspective baffling— think MuZero and LLMs are wildly different from an alignment perspective" isn't really related to this from my perspective. Like things ca... | 2025-07-05T18:49:15.635Z | 9 | YFwziCm2ewAtGYNXp | 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 |
wkoyi7M6iJrSBek4q | > I somehow completely agree with both of your perspectives, have you tried to ban the word "continuous" in your discussions yet?
I agree taboo-ing is a good approach in this sort of case. Talking about "continuous" wasn't a big part of my discussion with Steve, but I agree if it was. | 2025-07-05T21:00:29.437Z | 2 | CfvoX2jFRdLBDdBbF | 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 |
wFwxDChRjeLeBPgsz | Do reasoning models default to having good grammar in their thoughts? (Even when they are trained to have shorter reasoning.)
Here is an example of reasoning from "[Detecting misbehavior in frontier reasoning models](https://openai.com/index/chain-of-thought-monitoring/)":
> Now implement accordingly.
>
> Memo.reset(... | 2025-07-15T22:51:12.977Z | 13 | 8sDXB7XciCyFzmSKt | 7xneDbsgj6yJDJMjK | Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety | chain-of-thought-monitorability-a-new-and-fragile | https://bit.ly/cot-monitorability-fragile | Tomek Korbak | 2025-07-15T16:23:17.588Z |
vFohAqhrepsTWv6YD | > Anyway, this is my crux. If we start to see competent agentic behavior I will buy into the short timelines view at 75% +
Seems good to flesh out what you mean by this if it's such a big crux. Ideally, you'd be able to flesh this out in such a way that bad vision (a key problem for games like pokemon) and poor motiva... | 2025-07-16T18:41:49.457Z | 13 | D6GNooEYrbJNREf8E | 5tqFT3bcTekvico4d | Do confident short timelines make sense? | do-confident-short-timelines-make-sense | https://www.lesswrong.com/posts/5tqFT3bcTekvico4d/do-confident-short-timelines-make-sense | TsviBT | 2025-07-15T03:37:02.390Z |
eD7RASM5GyTmziNeu | > Adversarial robustness is part of agency, so I don't agree with that aspect of your framing.
Maybe so, but it isn't clearly required for automating AI R&D! | 2025-07-16T23:14:38.697Z | 2 | AMy8ZvkEydQendLmq | 5tqFT3bcTekvico4d | Do confident short timelines make sense? | do-confident-short-timelines-make-sense | https://www.lesswrong.com/posts/5tqFT3bcTekvico4d/do-confident-short-timelines-make-sense | TsviBT | 2025-07-15T03:37:02.390Z |
76xErMSCcdSammzxN | IMO, the type of adversarial robustness you're discussing is sufficiently different than what people typically mean by adversarial robustness that it would be worth tabooing the word. (E.g., I might say "robust self-verification is required".) | 2025-07-17T00:23:45.412Z | 8 | bM7gEQCTzGwNzpLN9 | 5tqFT3bcTekvico4d | Do confident short timelines make sense? | do-confident-short-timelines-make-sense | https://www.lesswrong.com/posts/5tqFT3bcTekvico4d/do-confident-short-timelines-make-sense | TsviBT | 2025-07-15T03:37:02.390Z |
mca28LA3f7Kh7DH5q | You come away with the conclusion that "I think the best futures at least would require a good deal of preventing constraining competition, at least re-locust like value systems, and this despite many risks that this entails."
I don't understand why you think competition with locusts probably burns much of the galacti... | 2025-07-17T23:10:26.568Z | 7 | null | evYne4Xx7L9J96BHW | Video and transcript of talk on "Can goodness compete?" | video-and-transcript-of-talk-on-can-goodness-compete | https://joecarlsmith.substack.com/p/video-and-transcript-of-talk-on-can | Joe Carlsmith | 2025-07-17T17:54:17.425Z |
pKjkGdodNFXRedcyh | Here are two other potentially serious failures of strategy stealing[^comp]:
- [Vacuum decay](https://www.lesswrong.com/posts/zteMisMhEjwhZbWez/vacuum-decay-expert-survey-results) might allow for agents who control small amounts of resources to destory everything if there isn't relevant policing/prevention everywhere.... | 2025-07-17T23:19:58.926Z | 6 | null | evYne4Xx7L9J96BHW | Video and transcript of talk on "Can goodness compete?" | video-and-transcript-of-talk-on-can-goodness-compete | https://joecarlsmith.substack.com/p/video-and-transcript-of-talk-on-can | Joe Carlsmith | 2025-07-17T17:54:17.425Z |
zhnuzzwyKsDmr9DPo | I think that "we should only slow down AI development if for each year of slowing down we would be reducing risk of human extinction by more than 1%" is not a sufficient crux for the (expensive) actions which I most want at current margins, at least if you have my empirical views. I think it is very unlikely (~7%?) tha... | 2025-07-18T00:32:05.777Z | 2 | gmdQgZs2BiBgP9CQa | CYTwRZtrhHuYf7QYu | A case for courage, when speaking of AI danger | a-case-for-courage-when-speaking-of-ai-danger | https://www.lesswrong.com/posts/CYTwRZtrhHuYf7QYu/a-case-for-courage-when-speaking-of-ai-danger | So8res | 2025-06-27T02:15:25.781Z |
MCYvKFx5EuMDmRYaA | > People who hold this position are arguing for things like "we should only slow down AI development if for each year of slowing down we would be reducing risk of human extinction by more than 1%", which is a policy that if acted on consistently would more likely than not cause humanity's extinction within 100 years (a... | 2025-07-18T00:46:12.439Z | 2 | gmdQgZs2BiBgP9CQa | CYTwRZtrhHuYf7QYu | A case for courage, when speaking of AI danger | a-case-for-courage-when-speaking-of-ai-danger | https://www.lesswrong.com/posts/CYTwRZtrhHuYf7QYu/a-case-for-courage-when-speaking-of-ai-danger | So8res | 2025-06-27T02:15:25.781Z |
7bSiL4KHoZxt7qw9Z | When I say space combat / exploration dynamics, I'm not including acausal interactions. I don't think that locusts are especially a problem for acausal interactions. | 2025-07-18T14:52:09.882Z | 2 | fuQAunocRPNmhsqGn | evYne4Xx7L9J96BHW | Video and transcript of talk on "Can goodness compete?" | video-and-transcript-of-talk-on-can-goodness-compete | https://joecarlsmith.substack.com/p/video-and-transcript-of-talk-on-can | Joe Carlsmith | 2025-07-17T17:54:17.425Z |
uLtdENqdGCKQbweDP | By "defense dominant", I meant "the person who gets to resources first can defend them from other actors who want these resources cheaply" (or something roughly like this). Vacuum decay gets no one any resources, so it feels natural to pull out into a separate category. I agree that value systems which want vacuum deca... | 2025-07-18T14:58:21.619Z | 4 | 4YttgNjSGWasW7BSg | evYne4Xx7L9J96BHW | Video and transcript of talk on "Can goodness compete?" | video-and-transcript-of-talk-on-can-goodness-compete | https://joecarlsmith.substack.com/p/video-and-transcript-of-talk-on-can | Joe Carlsmith | 2025-07-17T17:54:17.425Z |
EiivcsHbqbqqM349b | The short version containing some of my understanding: Things in space are far away and expensive to reach in a timely way. So, if someone gets to some resources first, they can spend a small fraction on defenses and be reasonably well defended unless an attacker spends far more resources sending over things to attack ... | 2025-07-18T17:10:20.046Z | 4 | fzTsr3bb3ydavan48 | evYne4Xx7L9J96BHW | Video and transcript of talk on "Can goodness compete?" | video-and-transcript-of-talk-on-can-goodness-compete | https://joecarlsmith.substack.com/p/video-and-transcript-of-talk-on-can | Joe Carlsmith | 2025-07-17T17:54:17.425Z |
d8SJkCwX5JtEa6SDt | Sure, I agree this is possible in principle. (And if many uncaring actors can cause vacuum decay then it makes sense that at least one of them might have a slight incentive to do this.) | 2025-07-18T17:57:48.192Z | 2 | i3iuRqvufiSk3bfvZ | evYne4Xx7L9J96BHW | Video and transcript of talk on "Can goodness compete?" | video-and-transcript-of-talk-on-can-goodness-compete | https://joecarlsmith.substack.com/p/video-and-transcript-of-talk-on-can | Joe Carlsmith | 2025-07-17T17:54:17.425Z |
ajToDqPvw53x7MHxh | > But since when are math proofs, or competitive programming, considered good examples of hard-to-verify tasks?
When Noam Brown says "hard-to-verify", I think he means that natural language IMO proofs are "substantially harder to verify": he says "proofs are pages long and take experts hours to grade".
(Yes, there ar... | 2025-07-21T16:55:48.455Z | 6 | ZpDqNS3dBB7Ti93dm | RcBqeJ8GHM2LygQK3 | OpenAI Claims IMO Gold Medal | openai-claims-imo-gold-medal | https://x.com/alexwei_/status/1946477742855532918 | Mikhail Samin | 2025-07-19T09:58:19.555Z |
3iq3PYyWTJizMWGjw | If the AI is iterating on solutions, there is actually pressure to reduce the length of draft/candidate solutions. Then, it might be that OpenAI didn't implement a clean up pass on the final solution (even though there wouldn't be any real pressure to save tokens in the final clean up). | 2025-07-21T17:39:09.067Z | 4 | 7dWLkGkp3xaq7gmZP | RcBqeJ8GHM2LygQK3 | OpenAI Claims IMO Gold Medal | openai-claims-imo-gold-medal | https://x.com/alexwei_/status/1946477742855532918 | Mikhail Samin | 2025-07-19T09:58:19.555Z |
X73ao6xbMR5A6jx3k | It's worth distinguishing between obfuscated CoT and steganographic CoT. If thinking out loud in the CoT is a necessity for a given AI for some problematic task/objective, then encoding the reasoning stenographically is likely[^likely] to be required to evade CoT monitoring. The results here don't correspond to the AI ... | 2025-07-30T22:19:04.501Z | 24 | null | CM7AsQoBxDW4vhkP3 | [Research Note] Optimizing The Final Output Can Obfuscate CoT | research-note-optimizing-the-final-output-can-obfuscate-cot | https://www.lesswrong.com/posts/CM7AsQoBxDW4vhkP3/research-note-optimizing-the-final-output-can-obfuscate-cot | lukemarks | 2025-07-30T21:26:42.796Z |
RJkKckjCFBd7MCk6M | Another type of ensembling which helps due to both uncertainty about scheming and a reduced chance of (successful) collusion is ensembling over AIs from multiple different providers as noted [here](https://www.lesswrong.com/posts/4WvHqPzaxAJXZrND7/cam-s-shortform?commentId=SFM9d2Lmk3gHKPRho). In practice, this might be... | 2025-08-03T23:29:36.699Z | 2 | XrENSNCXqWMwfek3f | GCqoks9eZDfpL8L3Q | How to prevent collusion when using untrusted models to monitor each other | how-to-prevent-collusion-when-using-untrusted-models-to | https://www.lesswrong.com/posts/GCqoks9eZDfpL8L3Q/how-to-prevent-collusion-when-using-untrusted-models-to | Buck | 2024-09-25T18:58:20.693Z |
mmouudcSjovCxe4cx | This seems pretty reasonable (strong upvoted).
These aren't disagreements with the post, but here are some of my thoughts (most of which Sam already knows, but I'm sharing for third parties):
My largest concern is that the auditing test beds will end up being pretty disanalogous to the cases of misalignment I find mo... | 2025-08-05T01:59:22.397Z | 14 | null | bGYQgBPEyHidnZCdE | Towards Alignment Auditing as a Numbers-Go-Up Science | towards-alignment-auditing-as-a-numbers-go-up-science | https://www.lesswrong.com/posts/bGYQgBPEyHidnZCdE/towards-alignment-auditing-as-a-numbers-go-up-science | Sam Marks | 2025-08-04T22:30:52.101Z |
fiqdNzjTabAjpxgrM | I comment some on the situation with CBRN/bio with respect to these models [here](https://www.lesswrong.com/posts/FG54euEAesRkSZuJN/ryan_greenblatt-s-shortform?commentId=xZQ4JpqPf68gwBq8L). | 2025-08-05T19:26:13.917Z | 14 | null | oPqgSrfkTt2sCKM3e | OpenAI releases gpt-oss | openai-releases-gpt-oss | https://openai.com/index/introducing-gpt-oss/ | anaguma | 2025-08-05T17:32:17.939Z |
jTh5oyAACvwK5fTrb | > As soon as AI can do X at all (or very soon afterwards), AI vastly outstrips any human’s ability to do X. This is a common enough pattern in AI, at this point, to barely warrant mentioning.
Citation needed? This seems false (thus far) in the case of Go, Chess, math, essay writing, writing code, recognizing if an ima... | 2025-08-06T01:30:19.835Z | 185 | null | kgb58RL88YChkkBNf | The Problem | the-problem | https://www.lesswrong.com/posts/kgb58RL88YChkkBNf/the-problem | Rob Bensinger | 2025-08-05T21:40:00.962Z |
ociPyQMgt23Xy9Wvm | Their post about honey seemed epistemically bad / misleading from my perspective, though maybe still interesting enough to be net positive on net.
This post seems basically fine to me. I'd also note that the argument in this post is basically the same as the argument made in [Torture vs. Dust Specks](https://www.lessw... | 2025-08-06T22:20:30.650Z | 3 | aqz5EJhz9a2jJB2kW | 7DXDovnfQ4dhupXeM | It's Better To Save Infinite Shrimp From Torture Than To Save One Person
| it-s-better-to-save-infinite-shrimp-from-torture-than-to | https://www.lesswrong.com/posts/7DXDovnfQ4dhupXeM/it-s-better-to-save-infinite-shrimp-from-torture-than-to | Bentham's Bulldog | 2025-08-05T16:10:30.538Z |
9ty25itqBFANbb7BE | Assuming this is talking about "[Alignment faking in large language models](https://www.lesswrong.com/posts/njAZwT8nkHnjipJku/alignment-faking-in-large-language-models)", the core prompting experiments were originally done by me (the lead author of the paper) and I'm not an Anthropic employee. So the main results can't... | 2025-08-06T23:16:44.261Z | 65 | null | oDX5vcDTEei8WuoBx | Re: recent Anthropic safety research | re-recent-anthropic-safety-research | https://x.com/ESYudkowsky/status/1952422379478741301 | Eliezer Yudkowsky | 2025-08-06T22:52:44.203Z |
dTQBanXAiJkQYWYJc | Sure, but the results would have come out either way. I agree the promotion could be a PR play. This generally limits the complexity of the PR play, but doen't rule out (e.g.) directional stuff. | 2025-08-06T23:26:18.695Z | 11 | fjkQicpka6B9g3opm | oDX5vcDTEei8WuoBx | Re: recent Anthropic safety research | re-recent-anthropic-safety-research | https://x.com/ESYudkowsky/status/1952422379478741301 | Eliezer Yudkowsky | 2025-08-06T22:52:44.203Z |
i3A7h6oLbYPGi359h | This comment is responding just to the paragraph arguing I'm wrong due to the context of "narrow domains". I feel like the arguments in that paragraph are very tenuous. This doesn't alter the rest of the point you're making, so I thought I would contain this in a separate comment.
Now, responding to the first part of ... | 2025-08-10T22:12:02.261Z | 7 | jo43rudGi9mpsSLef | kgb58RL88YChkkBNf | The Problem | the-problem | https://www.lesswrong.com/posts/kgb58RL88YChkkBNf/the-problem | Rob Bensinger | 2025-08-05T21:40:00.962Z |
M8y72ifdQgzdTCgtH | > Backing up, there is a basic point that I think The Problem is making, that I think is solid and I'm curious if you agree with. Paraphrasing: Many people underestimate the danger of superhuman AI because they mistakenly believe that skilled humans are close to the top of the range of mental ability in most domains.
... | 2025-08-10T22:45:26.865Z | 5 | jo43rudGi9mpsSLef | kgb58RL88YChkkBNf | The Problem | the-problem | https://www.lesswrong.com/posts/kgb58RL88YChkkBNf/the-problem | Rob Bensinger | 2025-08-05T21:40:00.962Z |
NiexKjYnzLWdW4Tsz | My view is that we can get a bunch of improvement in safety without massive shifts to the Overton window and poorly executed attempts at shifting the Overton window with bad argumentation (or bad optics) can poison other efforts.
I think well-executed attempts at massively shifting the Overton window are great and sho... | 2025-08-10T22:51:00.988Z | 6 | Qzw22SCj9ow78tyH4 | kgb58RL88YChkkBNf | The Problem | the-problem | https://www.lesswrong.com/posts/kgb58RL88YChkkBNf/the-problem | Rob Bensinger | 2025-08-05T21:40:00.962Z |
biNjmeHfeQEaowGbR | I see this post as trying to argue for a thesis that "if smarter-than-human AI is developed this decade, the result will be an unprecedented catastrophe." is true with reasonably high confidence and a (less emphasized) thesis that the best/only intervention is not building ASI for a long time: "The main way we see to a... | 2025-08-10T22:59:06.452Z | 22 | Xocu98ghGgaKLaHgi | kgb58RL88YChkkBNf | The Problem | the-problem | https://www.lesswrong.com/posts/kgb58RL88YChkkBNf/the-problem | Rob Bensinger | 2025-08-05T21:40:00.962Z |
3KGm7vjcHtftdg6gr | > then be nitpicked to death because not enough air was left in the room for an infinite sea of sub-cases of the fringe view, when leaving enough air in the room for them could require pushing the public out entirely, or undercut the message by priming the audience to expect that some geniuses off screen in fact just h... | 2025-08-11T03:19:54.685Z | 6 | ZoJXEGSkqFhGboSXq | kgb58RL88YChkkBNf | The Problem | the-problem | https://www.lesswrong.com/posts/kgb58RL88YChkkBNf/the-problem | Rob Bensinger | 2025-08-05T21:40:00.962Z |
X8dZ6zbiCEk4mEuHL | I don't really buy this doom is clearly the default frame. I'm not sure how important this is, but I thought I would express my perspective.
> But all of those stories look totally wild to me, and it's extremely difficult to see the mechanisms by which they might come to pass
A reasonable fraction of my non-doom worl... | 2025-08-11T03:32:48.974Z | 23 | ZoJXEGSkqFhGboSXq | kgb58RL88YChkkBNf | The Problem | the-problem | https://www.lesswrong.com/posts/kgb58RL88YChkkBNf/the-problem | Rob Bensinger | 2025-08-05T21:40:00.962Z |
atovvQQ94AaCuqp7H | > By what mechanism? This feels like 'we get a pause' or 'there's a wall'.
I just meant that takeoff isn't that fast so we have like >0.5-1 year at a point where AIs are at least very helpful for safety work (if reasonably elicited) which feels plausible to me. The duration of "AIs could fully automate safety (includi... | 2025-08-11T05:31:34.254Z | 8 | u9KQ5QCxzwnobkrB4 | kgb58RL88YChkkBNf | The Problem | the-problem | https://www.lesswrong.com/posts/kgb58RL88YChkkBNf/the-problem | Rob Bensinger | 2025-08-05T21:40:00.962Z |
cdvowJr7Hth8XAhDD | > More concretely, my median is that AI research will be automated by the end of 2028, and AI will be better than humans at >95% of current intellectual labor by the end of 2029.
It seems like you think AI research will be (fully?) automated once AIs are at 80% reliability on 1 month long benchmarkable / easy-to-check... | 2025-08-12T22:46:53.499Z | 11 | null | GAJbegsvnd85hX3eS | Thoughts on extrapolating time horizons | thoughts-on-extrapolating-time-horizons | https://x.com/nikolaj2030/status/1954248757513720297 | Nikola Jurkovic | 2025-08-11T22:36:35.033Z |
rL6DT6SNZhj6yP2PK | Suppose that we ended up with AIs that were ~perfectly aligned (to what the company/project that trained these AIs wanted) which had capabilities that dominate top human experts in AI R&D, AI safety research, philosophy, and most digital research fields in general. These AIs are sufficiently aligned that they try hard ... | 2025-08-13T01:36:08.144Z | 9 | dowuyzxhvSfcrjHoa | kgb58RL88YChkkBNf | The Problem | the-problem | https://www.lesswrong.com/posts/kgb58RL88YChkkBNf/the-problem | Rob Bensinger | 2025-08-05T21:40:00.962Z |
iXpamMBCtE8JRgRFe | Ok, sound it sounds like your view is "indeed if we got ~totally aligned AIs capable of fully automating safety work (but not notably more capable than the bare minimum requirement for this), we'd probably fine (even if there is still a small fraction of effort spent on safety) and the crux is earlier than this".
Is t... | 2025-08-14T01:04:28.323Z | 5 | kpptyWTFrfx3Re4dh | kgb58RL88YChkkBNf | The Problem | the-problem | https://www.lesswrong.com/posts/kgb58RL88YChkkBNf/the-problem | Rob Bensinger | 2025-08-05T21:40:00.962Z |
krMWsHLYuAPG3A7Dr | Are you looking at mean squared error in log space? I expect this will be more meaningful and less dominated by the last few points. | 2025-08-19T13:25:31.275Z | 3 | null | mXa66dPR8hmHgndP5 | Hyperbolic trend with upcoming singularity fits METR capabilities estimates. | hyperbolic-trend-with-upcoming-singularity-fits-metr | https://www.lesswrong.com/posts/mXa66dPR8hmHgndP5/hyperbolic-trend-with-upcoming-singularity-fits-metr | Valentin2026 | 2025-08-19T11:41:39.931Z |
4499W7vbZEcip8sq6 | From my perspective, the main case for working on safeguards is transfer to misalignment risk. (Both in terms of transferring of the methods and the policy/politics that this work maybe helps with transferring.)
I don't have a strong view about how much marginal work on the safeguards team transfers to misalignment ri... | 2025-08-21T04:48:39.756Z | 4 | P2BnANj8jtk8bJiYG | qBbnXtt9zFWXMrP4M | come work on dangerous capability mitigations at Anthropic | come-work-on-dangerous-capability-mitigations-at-anthropic | https://www.lesswrong.com/posts/qBbnXtt9zFWXMrP4M/come-work-on-dangerous-capability-mitigations-at-anthropic | Dave Orr | 2025-08-20T15:11:09.629Z |
gu99bhTJec9YrkZpk | > This rhymes with what Paul Christiano and his various interlocutors (e.g. Buck and Ryan above) think, but I think you've put forward a much weaker version of it than they do.
To be clear, I'm not at all expecting ASI to "self-align", "develop values that are benevolent towards us", or to pursue "cooperation, appease... | 2025-08-21T15:11:00.333Z | 5 | gFQ9L2z5kmjXhFbqE | kgb58RL88YChkkBNf | The Problem | the-problem | https://www.lesswrong.com/posts/kgb58RL88YChkkBNf/the-problem | Rob Bensinger | 2025-08-05T21:40:00.962Z |
EYjZ6GWzBPfhjAbnr | > Don't we feel gratitude and warmth and empathy and care-for-the-monkey's-values such that we're willing to make small sacrifices on their behalf?
People do make small sacrifices on behalf of monkeys? Like >1 / billion of human resources are spent on doing things for monkeys (this is just >$100k per year). And, in th... | 2025-08-21T15:32:28.021Z | 9 | EtQyzKcKDfyizX6zq | kgb58RL88YChkkBNf | The Problem | the-problem | https://www.lesswrong.com/posts/kgb58RL88YChkkBNf/the-problem | Rob Bensinger | 2025-08-05T21:40:00.962Z |
H4HLm9LpvXzJiJSAg | To be clear, Buck's view is that it is a very bad outcome if a token population is kept alive (e.g., all/most currently alive humans) but (misaligned) AIs control the vast majority of resources. And, he thinks most of the badness is due to the loss of the vast majority of resources.
He didn't say "and this would be fi... | 2025-08-21T15:35:24.096Z | 12 | 24xdXjXxMyQvsbfG5 | kgb58RL88YChkkBNf | The Problem | the-problem | https://www.lesswrong.com/posts/kgb58RL88YChkkBNf/the-problem | Rob Bensinger | 2025-08-05T21:40:00.962Z |
b9F9J3vxMrnHN574p | > But if this were true, you’d think they’d be able to handle ARC-AGI puzzles (see the example image just above)
In a footnote you note that models do well on ARC-AGI-1, but I think you're description of the situation is misleading:
* AIs trained on the training set of ARC-AGI (and given a bunch of compute) can *be... | 2025-08-27T20:56:39.806Z | 2 | null | uAbbEz4p6tcsENaRz | 35 Thoughts About AGI and 1 About GPT-5 | 35-thoughts-about-agi-and-1-about-gpt-5 | https://secondthoughts.ai/p/thoughts-about-agi-and-gpt-5 | snewman | 2025-08-16T19:20:17.672Z |
G8CwrsqbvJzv8EDiL | > Is sample-efficient learning a singularly important step on the path to AGI?
Almost definitionally, learning as efficiently as top humans would suffice for AGI. (You could just train the AI on way more data/compute and it would be superhuman.)
AIs will probably reach milestones like full automation of AI R&D before... | 2025-08-27T20:59:07.376Z | 3 | null | uAbbEz4p6tcsENaRz | 35 Thoughts About AGI and 1 About GPT-5 | 35-thoughts-about-agi-and-1-about-gpt-5 | https://secondthoughts.ai/p/thoughts-about-agi-and-gpt-5 | snewman | 2025-08-16T19:20:17.672Z |
4AckMBy3dgS6txZi8 | > AIs have been demonstrating what arguably constitutes superhuman performance on FrontierMath, a set of extremely difficult mathematical problems.
AIs aren't superhuman on frontier math. I'd guess that Terry Tao with 8 hours per problem (and internet access) is much better than current AIs. (Especially after practici... | 2025-08-27T21:02:44.491Z | 3 | null | uAbbEz4p6tcsENaRz | 35 Thoughts About AGI and 1 About GPT-5 | 35-thoughts-about-agi-and-1-about-gpt-5 | https://secondthoughts.ai/p/thoughts-about-agi-and-gpt-5 | snewman | 2025-08-16T19:20:17.672Z |
wiNa6JfLEB4sFxFzC | > To put it another way: compared to people, large language models seem to be superhuman in crystallized knowledge, which seems to be masking shortcomings in fluid intelligence. Is that a dead end, great for benchmarks but bad for a lot of work in the real world?
You seem to imply that AIs aren't improving on fluid in... | 2025-08-27T21:08:14.221Z | 4 | null | uAbbEz4p6tcsENaRz | 35 Thoughts About AGI and 1 About GPT-5 | 35-thoughts-about-agi-and-1-about-gpt-5 | https://secondthoughts.ai/p/thoughts-about-agi-and-gpt-5 | snewman | 2025-08-16T19:20:17.672Z |
LoqpxdbsbGz2NbjEK | > The best public estimate is that GPT-4 has 1.8 trillion “parameters”, meaning that its neural network has that many connections. In the two and a half years since it was released, it’s not clear that any larger models have been deployed (GPT-4.5 and Grok 3 might be somewhat larger).
>
> The human brain is far mo... | 2025-08-27T21:19:36.613Z | 6 | null | uAbbEz4p6tcsENaRz | 35 Thoughts About AGI and 1 About GPT-5 | 35-thoughts-about-agi-and-1-about-gpt-5 | https://secondthoughts.ai/p/thoughts-about-agi-and-gpt-5 | snewman | 2025-08-16T19:20:17.672Z |
QMiMFMAKqgnSweiQG | In general, this post seems to make a bunch of claims that LLMs have specific large qualitative barriers relative to humans, but these claims seem mostly unsupported. The evidence as far as I can tell seems more consistent with LLMs being weaker in a bunch of specific quantitative ways which are improving over time. Fo... | 2025-08-27T21:22:20.538Z | 3 | null | uAbbEz4p6tcsENaRz | 35 Thoughts About AGI and 1 About GPT-5 | 35-thoughts-about-agi-and-1-about-gpt-5 | https://secondthoughts.ai/p/thoughts-about-agi-and-gpt-5 | snewman | 2025-08-16T19:20:17.672Z |
qR6YjoGpMmxBsFD5t | [@Steven Byrnes](https://www.lesswrong.com/users/steve2152?mention=user) might have takes on this. | 2025-08-27T21:33:29.377Z | 2 | LoqpxdbsbGz2NbjEK | uAbbEz4p6tcsENaRz | 35 Thoughts About AGI and 1 About GPT-5 | 35-thoughts-about-agi-and-1-about-gpt-5 | https://secondthoughts.ai/p/thoughts-about-agi-and-gpt-5 | snewman | 2025-08-16T19:20:17.672Z |
W2HiDgz4abZXESqoW | > Say more? At https://arcprize.org/leaderboard, I see "Stem Grad" at 98% on ARC-AGI-1, and the highest listed AI score is 75.7% for "o3-preview (Low)". I vaguely recall seeing a higher reported figure somewhere for some AI model, but not 98%.
By "can beat humans", I mean AIs are well within the human range, probably ... | 2025-08-28T16:22:29.425Z | 4 | E4rfEPzneg7mdKx3c | uAbbEz4p6tcsENaRz | 35 Thoughts About AGI and 1 About GPT-5 | 35-thoughts-about-agi-and-1-about-gpt-5 | https://secondthoughts.ai/p/thoughts-about-agi-and-gpt-5 | snewman | 2025-08-16T19:20:17.672Z |
YPxQWJ8rgvafgg5JP | > Anyway, I think we may be agreeing on the main point here: my suggestion that LLMs solve FrontierMath problems "the wrong way", and your point about depth arguably being more important than breadth, seem to be pointing in the same direction.
Yep, though it's worth distinguishing between LLMs *often* solving Frontier... | 2025-08-28T16:25:32.186Z | 3 | 5yT5JhdQhqddcbfv9 | uAbbEz4p6tcsENaRz | 35 Thoughts About AGI and 1 About GPT-5 | 35-thoughts-about-agi-and-1-about-gpt-5 | https://secondthoughts.ai/p/thoughts-about-agi-and-gpt-5 | snewman | 2025-08-16T19:20:17.672Z |
rvtbAunGpsjfkouDe | > It just suggests (to me) that they might be farther from that milestone than their performance on crystal-intelligence-friendly tasks would imply.
I basically agree, but we can more directly attempt extrapolations (e.g. METR horizon length) and I put more weight on this.
I also find it a bit silly when people say "... | 2025-08-28T16:28:24.932Z | 2 | jmDfbmthkyhCHvWd2 | uAbbEz4p6tcsENaRz | 35 Thoughts About AGI and 1 About GPT-5 | 35-thoughts-about-agi-and-1-about-gpt-5 | https://secondthoughts.ai/p/thoughts-about-agi-and-gpt-5 | snewman | 2025-08-16T19:20:17.672Z |
Qjjrtsk37G4FRNzjF | > Regarding GPT-4, I believe the estimate was that it has 1.8 trillion parameters, which if shared weights are used may not precisely correspond to connections or FLOPs.
For standard LLM architectures, forward pass FLOPs are $\approx 2 \cdot \mathrm{parameters}$ (because of the multiply and accumulate for each matmul ... | 2025-08-28T17:57:00.546Z | 2 | 9ZhJa92gYnbT4FBph | uAbbEz4p6tcsENaRz | 35 Thoughts About AGI and 1 About GPT-5 | 35-thoughts-about-agi-and-1-about-gpt-5 | https://secondthoughts.ai/p/thoughts-about-agi-and-gpt-5 | snewman | 2025-08-16T19:20:17.672Z |
GSmDKjaWtn6ukefrG | Note that this post is arguing that there are some specific epistemic advantages of working as a moderate, not that moderates are always correct or that there aren't epistemic disadvantages to being a moderate. I don't think "there exist moderates which seem very incorrect to me" is a valid response to the post similar... | 2025-08-29T15:46:04.045Z | 23 | AudP8djw4goyjzCie | 9MaTnw5sWeQrggYBG | An epistemic advantage of working as a moderate | an-epistemic-advantage-of-working-as-a-moderate | https://www.lesswrong.com/posts/9MaTnw5sWeQrggYBG/an-epistemic-advantage-of-working-as-a-moderate | Buck | 2025-08-20T17:47:01.026Z |
9xFftQqJFT4v7cZLS | This seems very wrong to me from my experience in 2022 (though maybe the situation was very different in 2021? Or maybe there is some other social circle that I wasn't exposed to which had these properties?). | 2025-08-29T15:47:54.638Z | 1 | u9ajvjNKEqaavYghL | 9MaTnw5sWeQrggYBG | An epistemic advantage of working as a moderate | an-epistemic-advantage-of-working-as-a-moderate | https://www.lesswrong.com/posts/9MaTnw5sWeQrggYBG/an-epistemic-advantage-of-working-as-a-moderate | Buck | 2025-08-20T17:47:01.026Z |
PcEzyXPibLyzLhPDk | > Carlsmith's Multiple Stage Fallacy risk estimate of 5% that involved only an 80% chance anyone would even try to build agentic AI?
This is false as stated. The report says:
The corresponding footnote ... | 2025-08-29T15:56:37.133Z | 33 | AudP8djw4goyjzCie | 9MaTnw5sWeQrggYBG | An epistemic advantage of working as a moderate | an-epistemic-advantage-of-working-as-a-moderate | https://www.lesswrong.com/posts/9MaTnw5sWeQrggYBG/an-epistemic-advantage-of-working-as-a-moderate | Buck | 2025-08-20T17:47:01.026Z |
Zg7RNyF9iQrm3qQs5 | I think williawa's characterization of how people reacted to bioanchors basically matches my experience and I'm skeptical of the claim that OpenPhil was very politically controlling of EA with respect to timelines.
And, I agree with the claim that that Eliezer often implies people interpreted bioanchors in some way th... | 2025-08-29T16:01:54.953Z | 3 | zznnKPKZsHisLqeLE | 9MaTnw5sWeQrggYBG | An epistemic advantage of working as a moderate | an-epistemic-advantage-of-working-as-a-moderate | https://www.lesswrong.com/posts/9MaTnw5sWeQrggYBG/an-epistemic-advantage-of-working-as-a-moderate | Buck | 2025-08-20T17:47:01.026Z |
zvd2AygxkCNSZrf8F | I agree with this in principle, but think that doing a good job of noting major failings of prominent moderates in the current environment would look very different than Eliezer's comment and requires something stronger than just giving examples of some moderates which seem incorrect to Eliezer.
Another way to put thi... | 2025-08-29T16:24:09.095Z | 16 | CiRNK66CJEbmZyZXP | 9MaTnw5sWeQrggYBG | An epistemic advantage of working as a moderate | an-epistemic-advantage-of-working-as-a-moderate | https://www.lesswrong.com/posts/9MaTnw5sWeQrggYBG/an-epistemic-advantage-of-working-as-a-moderate | Buck | 2025-08-20T17:47:01.026Z |
dqHpeQFsnhRSz6XAt | I wasn't talking in retrospect, but I meant something might larger than $1 billion by strong incentives and I really mean very specifically APS systems at the time when they are feasible to build.
The 10% would come from other approaches/architectures ending up being surprisingly better at the point when people could ... | 2025-08-29T18:07:05.475Z | 3 | XzTS9Mt2FSvjYu6ik | 9MaTnw5sWeQrggYBG | An epistemic advantage of working as a moderate | an-epistemic-advantage-of-working-as-a-moderate | https://www.lesswrong.com/posts/9MaTnw5sWeQrggYBG/an-epistemic-advantage-of-working-as-a-moderate | Buck | 2025-08-20T17:47:01.026Z |
jtgKFFPsRTbobF328 | > On a more object level, my main critique of the post is that almost all of the bullet points are even more true of, say, working as a physicist.
But (in the language of the post) both moderates and radicals are working in the epistemic domain not some unrelated domain. It's not that moderates and radicals are trying... | 2025-08-30T01:15:23.938Z | 2 | z6mZR4zjaTXApHChG | 9MaTnw5sWeQrggYBG | An epistemic advantage of working as a moderate | an-epistemic-advantage-of-working-as-a-moderate | https://www.lesswrong.com/posts/9MaTnw5sWeQrggYBG/an-epistemic-advantage-of-working-as-a-moderate | Buck | 2025-08-20T17:47:01.026Z |
LtBizcFB2rFTrmExH | Hmm, I think what I said was about half wrong and I want to retract my point.
That said, I think much of the relevant questions are overlapping (like, "how do we expect the future to generally go?", "why/how is AI risky?", "how fast will algorithmic progress go at various points?) and I interpret this post as just tal... | 2025-08-31T14:18:03.619Z | 6 | 5er28NDwBMPcwi97D | 9MaTnw5sWeQrggYBG | An epistemic advantage of working as a moderate | an-epistemic-advantage-of-working-as-a-moderate | https://www.lesswrong.com/posts/9MaTnw5sWeQrggYBG/an-epistemic-advantage-of-working-as-a-moderate | Buck | 2025-08-20T17:47:01.026Z |
nXCSkxsBkGKF57c88 | [Feel free not to respond / react with the "not worth getting into" emoji"]
> Severe or willful violation of our RSP, or misleading the public about it.
Should this be read as "[Severe or willful violation of our RSP] or [misleading the public about the RSP]" or should this be read as "[Severe or willful violation of... | 2025-09-03T21:55:30.488Z | 46 | wAZPZDRFe4om5EBAL | PBd7xPAh22y66rbme | Anthropic's leading researchers acted as moderate accelerationists | anthropic-s-leading-researchers-acted-as-moderate | https://www.lesswrong.com/posts/PBd7xPAh22y66rbme/anthropic-s-leading-researchers-acted-as-moderate | Remmelt | 2025-09-01T23:23:43.993Z |
fK4KEyDv4YxzoQQnz | I certainly agree that the LTBT has de jure control (or as you say "formal control").
By "strong control" I meant more precisely something like: "lots of influence in practice, e.g. the influence of Anthropic leadership is comparable to the influence that the LTBT itself is exerting in practice over appointments or co... | 2025-09-04T15:10:33.804Z | 68 | kCSbLieAKrwCqJ4pi | PBd7xPAh22y66rbme | Anthropic's leading researchers acted as moderate accelerationists | anthropic-s-leading-researchers-acted-as-moderate | https://www.lesswrong.com/posts/PBd7xPAh22y66rbme/anthropic-s-leading-researchers-acted-as-moderate | Remmelt | 2025-09-01T23:23:43.993Z |
gYMjeLbuntdHeuaZw | > The Time article is materially wrong about a bunch of stuff
Agreed which is why I noted this in my comment.[^foot] I think it's a bad sign that Anthropic seemingly actively sought out an article that ended up being wrong/misleading in a way which was convenient for Anthropic at the time and then didn't correct it.
... | 2025-09-04T19:49:26.959Z | 44 | Y8tinYT65HjFaJ2tz | PBd7xPAh22y66rbme | Anthropic's leading researchers acted as moderate accelerationists | anthropic-s-leading-researchers-acted-as-moderate | https://www.lesswrong.com/posts/PBd7xPAh22y66rbme/anthropic-s-leading-researchers-acted-as-moderate | Remmelt | 2025-09-01T23:23:43.993Z |
yHLNvYtfXi6dY5EZp | See also [To be legible, evidence of misalignment probably has to be behavioral](https://www.lesswrong.com/posts/4QRvFCzhFbedmNfp4/to-be-legible-evidence-of-misalignment-probably-has-to-be). | 2025-09-06T02:26:28.613Z | 3 | KueCfhKrbPeesNbah | DBn83cvA6PDeq8o5x | Interpretability is the best path to alignment | interpretability-is-the-best-path-to-alignment | https://www.lesswrong.com/posts/DBn83cvA6PDeq8o5x/interpretability-is-the-best-path-to-alignment | Arch223 | 2025-09-05T04:37:17.670Z |
RzqspzrDmPuaxwbnM | > I'm quite sympathetic to something like the caveated version of the title
Presumably, another problem with your caveated version of the title is that you don't expect literally everyone to die (at least not with high confidence) even if AIs take over. | 2025-09-17T15:14:18.624Z | 2 | null | P4xeb3jnFAYDdEEXs | I enjoyed most of IABIED | i-enjoyed-most-of-iabied | https://www.lesswrong.com/posts/P4xeb3jnFAYDdEEXs/i-enjoyed-most-of-iabied | Buck | 2025-09-17T04:34:52.540Z |
TDfAFxCcHkZ5LCgc3 | > I have never heard of a remotely realistic-seeming story for how things will be OK, without something that looks like coordination to not build ASI for quite a while.
I wonder if we should talk about this at some point. This perspective feels pretty wild to me and I don't immediately understand where the crux lives... | 2025-09-19T17:14:05.756Z | 29 | 48bEwnxJJy5XR7Lsk | P4xeb3jnFAYDdEEXs | I enjoyed most of IABIED | i-enjoyed-most-of-iabied | https://www.lesswrong.com/posts/P4xeb3jnFAYDdEEXs/i-enjoyed-most-of-iabied | Buck | 2025-09-17T04:34:52.540Z |
4JvznSDb9TLEgTo9L | Sure, several. E.g.:
- USG cares a decent amount and leading AI companies are on board, so they try to buy several additional years to work on safety.
- We scale to roughly top human expert level while ensuring control.
- Over time, we lower the risk of scheming at this level of capability through a bunch of empirical... | 2025-09-19T18:05:31.394Z | 6 | iASXwYJdjZszBgahh | P4xeb3jnFAYDdEEXs | I enjoyed most of IABIED | i-enjoyed-most-of-iabied | https://www.lesswrong.com/posts/P4xeb3jnFAYDdEEXs/i-enjoyed-most-of-iabied | Buck | 2025-09-17T04:34:52.540Z |
H9rRrS4zJAzg7xJPG | > I think at the very least you need to intervene to stop that feedback loop.
There's probably at least some disagreement here. I think even if you let takeoff proceed at the default rate with a small fraction (e.g. 5%) explicitly spent on reasonably targeted alignment work at each point (as in, 5% beyond what is pure... | 2025-09-19T23:48:49.993Z | 10 | g5FTYAzbEjnuNy3Rm | P4xeb3jnFAYDdEEXs | I enjoyed most of IABIED | i-enjoyed-most-of-iabied | https://www.lesswrong.com/posts/P4xeb3jnFAYDdEEXs/i-enjoyed-most-of-iabied | Buck | 2025-09-17T04:34:52.540Z |
QJiZyuZ87ovhqgq5X | > Your first paragraph is an example of "something that looks like coordination to not build ASI for quite a while"! "Several additional years" is definitely "quite a while"!
Ok, if you count a several additional years as quite a while, then we're probably closer to agreement.
---
For this scenario, I was imagining... | 2025-09-20T00:02:03.141Z | 5 | 8jG4ipDpgkEspq7W6 | P4xeb3jnFAYDdEEXs | I enjoyed most of IABIED | i-enjoyed-most-of-iabied | https://www.lesswrong.com/posts/P4xeb3jnFAYDdEEXs/i-enjoyed-most-of-iabied | Buck | 2025-09-17T04:34:52.540Z |
LHD3mum4imKvxbSvB | I think this comment is failing to engage with Rohin's perspective.
Rohin's claim presumably isn't that people shouldn't say anything that happens to be sensationalist, but instead that LW group epistemics have a huge issue with sensationalism bias.
> There are like 5 x-risk-scene-people I can think offhand who seem ... | 2025-09-20T16:23:37.933Z | 29 | forxGaAcjf9s4TH45 | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
Qx2sDtbgdC3qcmNiN | > The book is making a (relatively) narrow claim.
>
> You might *still* disagree with that claim. I think there are valid reasons to disagree, or at least assign significantly less confidence to the claim.
>
> But none of the reasons listed so far are disagreements with the thesis. And, remember, if the reason you ... | 2025-09-20T16:42:39.557Z | 41 | null | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
fZ5fz6BSnquBGxQJE | > Given the [counteraguments](https://ifanyonebuildsit.com/5/wont-ais-care-at-least-a-little-about-humans), I don't see a reason to think this more than single-digit-percent likely to be especially relevant. (I can see >9% likelihood the AIs are "nice enough that something interesting-ish happens" but not >9% likelihoo... | 2025-09-20T17:06:14.337Z | 18 | null | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
FKLfPSLWrmDZddDdg | > I find it very strange that Collier claims that international compute monitoring would “tank the global economy.” What is the mechanism for this, exactly?
>10% of the value of the S&P 500 is downstream of AI and the proposal is to ban further AI development. AI investment is a non-trivial fraction of current US GDP ... | 2025-09-20T17:15:57.999Z | 14 | null | JWH63Aed3TA2cTFMt | Contra Collier on IABIED | contra-collier-on-iabied | https://www.lesswrong.com/posts/JWH63Aed3TA2cTFMt/contra-collier-on-iabied | Max Harms | 2025-09-20T15:55:06.201Z |
QqCkofRQ7kRXQqJBf | > I... really don't know what Scott expected a story that featured actual superintelligence to look like. I think the authors bent over backwards giving us one of the least-sci-fi stories you could possibly tell that includes superintelligence doing anything at all, without resorting to "superintelligence just won't ev... | 2025-09-20T17:35:36.658Z | 8 | null | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
YWfDQo35Ww5yKcBdj | (Maybe this is obvious, but I thought I would say this just to be clear.)
> I think "anything like current techniques" and "anything like current understanding" clearly set a very high bar for the difference. "We made more progress on interpretability/etc at the current rates of progress" fairly clearly doesn't count ... | 2025-09-21T16:38:23.507Z | 6 | Syc7j6nwbA6pxXZGw | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
SeePRjfZp2c2piZoj | To clarify my views on "will misaligned AIs that succeed in seizing all power have a reasonable chance of keeping (most/many) humans alive":
I think this isn't very decision relevant and is not that important. I think AI takeover kills the majority of humans *in expectation* due to both the takeover itself and killing... | 2025-09-21T17:38:56.230Z | 13 | hfMfey9waritrCm5L | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
ZrfFwZN57vGiRYfZ7 | I would say my summary for hope is more like:
- It seems pretty likely to be doable (with lots of human-directed weak AI labor and/or controlled stronger AI labor) to use iterative and prosaic methods within roughly the current paradigm to sufficiently align AIs which are slightly superhuman. In particular, AIs which ... | 2025-09-21T17:50:08.693Z | 7 | Q2fuLCxr6KA35oJHZ | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
hcPkhjDyoEzaeqnu6 | FWIW, I don't really consider my self to be responding to the book at all (in a way that is public or salient to your relevant audience) and my reasons for not signal boosting the book aren't really downstream of the content in the book in the way you describe. (More like, I feel sign uncertain about making You/Eliezer... | 2025-09-21T17:55:18.831Z | 5 | FeyHd6WxHPkXzw7gN | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
mYTvwvbjNFS5ReHuz | > (b) it sounds like I'm willing to classify more things as "death" than you are.
I don't think this matters much. I'm happy to consider non-consensual uploading to be death and I'm certainly happy to consider "the humans are modified in some way they would find horrifying (at least on reflection)" to be death. I thin... | 2025-09-21T18:05:44.705Z | 12 | eqwxjdDyXrJWNuqnA | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
mKumcvvr45bqpDdPL | Why do I think the story involves a lot of discontinuity (relative to what I expect)?
- Right at the start of the story, Sable has much higher levels of capability than Galvanic expects. It can confortably prove the Riemann Hypothesis even though Galvanic engineers are impressed by it proving some modest theorems. Gen... | 2025-09-21T22:19:52.402Z | 17 | iJ5HvK2oHrequa3ZA | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
gbw3dik8x8WwGonmL | (I mean, I think it's a substantial reason to think that "literally everyone dies" is considerably less likely and makes me not want to say stuff like "everyone dies", but I just don't think it implies much optimism exactly because the chance of death still seems pretty high and the value of the future is still lost. L... | 2025-09-22T01:11:30.544Z | 17 | c4XZeqtyjyLFjhCa4 | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
XqfMauRhpr4kCSGmP | > I mean, I also believe that if we solve the alignment problem, then we will no longer have an alignment problem, and I predict the same is true of Nate and Eliezer.
By "superintelligence" I mean "systems which are qualititatively much smarter than top human experts". (If Anyone Builds It, Everyone Dies seems to defi... | 2025-09-22T01:22:52.969Z | 18 | B7pbKgE3W7Ha3qdQG | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
MQndsjXejBB3K9HYF | Thanks for the nudge! I currently disagree with "very unlikely", but more importantly, I noticed that I haven't really properly analyzed the question of "given how much cognitive labor is available between different capability levels, should we expect that alignment can keep up with capabilities if a small fraction (e.... | 2025-09-22T02:29:19.912Z | 11 | AfFnAn67npdfBzL7W | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
tHPtEhyYa7JbD6Rr8 | > I was glad to have read it, but am a bit confused about your numbers seeming different from the ones I objected to.
I gave "1 million AIs somewhat smarter than humans with the equivalent of 100 years each" as an example of a situation I thought wouldn't count as "anything like current techniques/understanding". In t... | 2025-09-22T14:56:51.561Z | 2 | pn5ixLBKB2zeLEfae | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
T9ehPWmbPxxSDoKvk | > The key disanalogy between capabilities and alignment work at the point of DAI is that the DAI might be scheming, but you're in a subjunctive case where we've assumed the DAI is not scheming. Whence the pessimism?
I don't think this is the only disanalogy. It seems to me like getting AIs to work efficiently on auto... | 2025-09-22T15:01:21.667Z | 4 | ppfeGpRBJk3t7u2qF | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
jZDfWEJAbEy8ggaSE | I don't think I understand this comment.
It sounds like you're saying:
"Slower takeoff should be correlated with 'harder' alignment (in terms of cognitive labor requirements) because slower takeoff implies returns to cognitive labor in capabilities R&D are relatively lower and we should expect this means that alignme... | 2025-09-22T15:09:42.456Z | 4 | pDLNx376XPTafxmen | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
KG9ikcvqabhJDM8YG | Do you agree the feedback loops for capabilities are better right now?
> Sure, AI societies could go off the rails that hurts alignment R&D but not AI R&D; they could also go off the rails in a way that hurts AI R&D and not alignment R&D. Not sure why I should expect one rather than the other.
For this argument it's ... | 2025-09-22T16:55:52.787Z | 2 | faAkLKRAt8dr2dNdE | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
EP2DuyyXG4MmHRkBY | > Yes, primarily due to the asymmetry where capabilities can work with existing systems while alignment is mostly stuck waiting for future systems, but that should be much less true by the time of DAI.
I think studying scheming in current/future systems has ongoingly worse feedback loops? Like suppose our DAI level sy... | 2025-09-22T18:57:13.046Z | 4 | XfrfwJrDKrvnfHEB8 | voEAJ9nFBAqau8pNN | The title is reasonable | the-title-is-reasonable | https://www.lesswrong.com/posts/voEAJ9nFBAqau8pNN/the-title-is-reasonable | Raemon | 2025-09-20T08:59:10.085Z |
pLqBXfP9WdwLhWaTg | > I can't actually think of that many cases of humans failing at aligning existing systems because the problem is too technically hard.
You're probably already tracking this, but the biggest cases of "alignment was actually pretty tricky" I'm aware are:
- Recent systems doing egregious reward hacking in some cases (i... | 2025-09-23T00:06:52.188Z | 14 | keZcn2ynewoXSqukp | JWH63Aed3TA2cTFMt | Contra Collier on IABIED | contra-collier-on-iabied | https://www.lesswrong.com/posts/JWH63Aed3TA2cTFMt/contra-collier-on-iabied | Max Harms | 2025-09-20T15:55:06.201Z |
R2uSFn3cFgKvL3ovR | > But, I do think it'd be good for someone to write a really fleshed out takeoff story and/or forecast that runs with those assumptions.
You might be interested in [this episode of Epoch After Hours](https://epoch.ai/epoch-after-hours/ai-in-2040). I think this is pretty close to what you want. | 2025-09-23T03:07:39.978Z | 5 | null | Xp9ie6pEWFT8Nnhka | Accelerando as a "Slow, Reasonably Nice Takeoff" Story | accelerando-as-a-slow-reasonably-nice-takeoff-story | https://www.lesswrong.com/posts/Xp9ie6pEWFT8Nnhka/accelerando-as-a-slow-reasonably-nice-takeoff-story | Raemon | 2025-09-23T02:15:51.854Z |
RBKLK9H4KGqRtTNCe | My best guess is that the elicitation overhang has reduced or remained stable after the release of GPT-4. I think we've now elicited much more of the underlying "easy to access" capabilities with RL, and there is a pretty good a priori case that scaled up RL should do an OK job eliciting capabilities (at least when con... | 2025-09-23T16:17:58.683Z | 40 | null | 4YvSSKTPhPC43K3vn | We are likely in an AI overhang, and this is bad. | we-are-likely-in-an-ai-overhang-and-this-is-bad | https://cognition.cafe/p/we-are-likely-in-an-ai-overhang-and | Gabriel Alfour | 2025-09-23T14:15:10.510Z |
gCF577fctD4psZZfs | I [responsed on x/twitter](https://x.com/RyanPGreenblatt/status/1972056421501124892). My response is somewhat redundant with what [habryka said](https://www.lesswrong.com/posts/iFRrJfkXEpR4hFcEv/a-reply-to-macaskill-on-if-anyone-builds-it-everyone-dies?commentId=AsoStwvpBmEpZ8EkF), but it seemed reasonable to copy it h... | 2025-09-28T03:42:32.283Z | 12 | null | iFRrJfkXEpR4hFcEv | A Reply to MacAskill on "If Anyone Builds It, Everyone Dies" | a-reply-to-macaskill-on-if-anyone-builds-it-everyone-dies | https://www.lesswrong.com/posts/iFRrJfkXEpR4hFcEv/a-reply-to-macaskill-on-if-anyone-builds-it-everyone-dies | Rob Bensinger | 2025-09-27T23:03:33.334Z |
BirhkDMJkwCPMAzXf | > I don't think it would be the case that model M_n develops the propensity to act covertly and pursue its own goals, but the only goal it cares about is taking over the world, and also identifies with future models, and so it decides to "lie in wait" until generation M_{n+k} where it would act on that.
Suppose that ... | 2025-09-29T16:08:48.001Z | 45 | TRbpQbsvfW4nD6MLC | CScshtFrSwwjWyP2m | A non-review of "If Anyone Builds It, Everyone Dies" | a-non-review-of-if-anyone-builds-it-everyone-dies | https://www.lesswrong.com/posts/CScshtFrSwwjWyP2m/a-non-review-of-if-anyone-builds-it-everyone-dies | Boaz Barak | 2025-09-28T17:34:17.782Z |
AuKnZmsjLz2arHbgw | To clarify my current perspective: I don't think current LLMs (released as of Oct 2025) plot against us or intentionally sabotage work to achieve their longer-run aims (for the vast, vast majority of typical usage; you can construct situations where they do this). I'd guess this isn't controversial. Correspondingly, I ... | 2025-10-07T01:16:45.893Z | 13 | null | QdEHS9TmBBKirz7Ky | LLMs are badly misaligned | llms-are-badly-misaligned | https://www.lesswrong.com/posts/QdEHS9TmBBKirz7Ky/llms-are-badly-misaligned | Joe Rogero | 2025-10-05T14:00:41.587Z |
pgaCb5nY8rKhkANgi | I clarify my current perspective [here](https://www.lesswrong.com/posts/QdEHS9TmBBKirz7Ky/llms-are-badly-misaligned?commentId=AuKnZmsjLz2arHbgw). | 2025-10-07T01:17:08.407Z | 2 | pJqTPoLWgoy5qbnLu | QdEHS9TmBBKirz7Ky | LLMs are badly misaligned | llms-are-badly-misaligned | https://www.lesswrong.com/posts/QdEHS9TmBBKirz7Ky/llms-are-badly-misaligned | Joe Rogero | 2025-10-05T14:00:41.587Z |
nBuKbiYMyngurkzE7 | > (I also acknowledge that I failed to parse what Ryan is saying in the parenthetical — does he mean that he mostly expects such research-capable AIs to be wildly superhuman, or that he mostly doesn’t? I infer from context that he doesn’t, but this is a guess.)
When I said "Let's say these AIs aren't much more capabl... | 2025-10-07T01:27:00.055Z | 4 | null | 8buEtNxCScYpjzgW8 | We won’t get AIs smart enough to solve alignment but too dumb to rebel | we-won-t-get-ais-smart-enough-to-solve-alignment-but-too | https://www.lesswrong.com/posts/8buEtNxCScYpjzgW8/we-won-t-get-ais-smart-enough-to-solve-alignment-but-too | Joe Rogero | 2025-10-06T21:49:05.595Z |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.