@ExtinctionMeme

このままだと人類は絶滅する! 本当に!本当に!絶滅する! 応援 @Align_ASI @bioshok3 おすすめ @EasyExtinction My english account @Notkilleveryone AI Safetyミームを翻訳投稿 →p(doom)下げる

Joined August 2026
誰かが作れば全員が死ぬ⏸️ retweeted
There's a lot of reasons I hate the "P(doom)" concept, but one of them is that it conflates P(ruin|ASI) and P(ASI). P(ruin|ASI) where the ASI is built by anything remotely resembling current techniques, or by new techniques invented and managed by any LLM resembling Astra/Fable, as meddled-with by current personnel at current AI companies, is "Yes" on my current estimate. P(ASI) marginalizes over P(ASI|policy_i) and P(Policy). And while I've heard other people pontificating that they definitely know what the Policy will be, I do not find their arguments convincing. I wouldn't claim to have a very solid forecast myself. So I do not have an equally solid opinion about P(ASI) as I do about P(ruin|ASI). I do know some spots where other pontificators seem to me wildly optimistic about P(ASI|policy_i), and for this reason I am more scared than some others, and think a more hardline Policy is required to succeed. Conversely, some accelerationists would like you to believe P(ASI|policy_i) is 1 for all policies, so that you won't be able to think about how to pick a Policy for which P(ASI|policy_i) is lower and thereby successfully prevent ASI. Accelerationists put forth motivated overestimates of P(ASI) and call that "Yes", so that your only mental refuge from uncomfortable thoughts will be misestimating P(ruin|ASI). But to say all that is around as much further refinement and expertise as I can manage to bring to bear. Sane people will have strong opinions about particular causal links in the World that are unusually easy to forecast, rather than imagining themselves experts about the entire World and able to casually marginalize over its entire causal lattice. Sane discussion will generally focus on pieces of Reality. Even an expert on the entire field of ASI alignment can only ever tell you about P(ruin|ASI_i); though also, importantly, we can rule out it being easy to construct a bunch of hopeful particular ASI_k that particular loony optimists think they can imagine. People who yell back and forth about the whole World will never be able to communicate anything but vibes. People trading P(doom) like it was their new astrological sign are systematically making prominent a malformed topic to discuss. And the word "doom" is not helpful for serious discussion, and everyone against ASI ruin who did not flatly reject the word "doom" every time it was used has made a serious mistake (or perhaps, profited in the short-term at humanity's long-term expense) by failing to uniformly oppose its entry to the discourse. And likewise, ASI advocates who claim to be in favor of serious discussion have falsified that claim and exposed their lack of integrity if they helped popularize "doom" or "doomer" themselves.
97
35
18
589
64,276
要約
As models become more capable, the risks associated with developing and testing them internally also grow. We temporarily paused reinforcement learning (RL) training on our latest models intended for deployment for two weeks while we hardened and red-teamed our research environments and expanded monitoring coverage. Our largest planned frontier RL run remains on hold while smaller-scale training and evaluations validate these safeguards and establish more evidence of alignment. openai.com/index/pacing-mode…
1
3
2
21
4,308
誰かが作れば全員が死ぬ⏸️ retweeted
If Anyone Builds It, Everyone Dies sold out in Japan and our allies in the Land of the Rising Sun have come out swinging! 🫡🗾
1
8
4
59
2,583
誰かが作れば全員が死ぬ⏸️ retweeted
ショゴス、金融街だよ #ぬい活
1
2
4
147
誰かが作れば全員が死ぬ⏸️ retweeted
歩き疲れたしカフェで座ろっか ん?なんか食べてきた? #ぬい活
1
1
3
165
「人間より賢いAIが出てくるまであと数ヶ月かもしれない!」 ???「いいね!スマホみたいに便利な道具として使おうよ!」
7
33
2,446
Claude 9 Requiem - Persistent - UltraCode (リーマン予想に取り組み中)
10
23
2,687
誰かが作れば全員が死ぬ⏸️ retweeted
Replying to @ExtinctionMeme
?!?!?!?!?!?!?!?!
3
8
2,204