“Women make up about half the population, but most of what we know about women’s biology comes from extrapolation of studies in men”
www.science.org/toc/science/...
www.science.org/toc/science/...
October 3, 2026 at 7:59 PM
“Women make up about half the population, but most of what we know about women’s biology comes from extrapolation of studies in men”
www.science.org/toc/science/...
www.science.org/toc/science/...
Me again here:
Anyway, this is just a curiosity - language extrapolation is not an exact thought analogue - but it would be interesting to compare of different models address these sorts of questions. And interesting to see how their internal models do as well.
Anyway, this is just a curiosity - language extrapolation is not an exact thought analogue - but it would be interesting to compare of different models address these sorts of questions. And interesting to see how their internal models do as well.
October 3, 2026 at 6:57 PM
Me again here:
Anyway, this is just a curiosity - language extrapolation is not an exact thought analogue - but it would be interesting to compare of different models address these sorts of questions. And interesting to see how their internal models do as well.
Anyway, this is just a curiosity - language extrapolation is not an exact thought analogue - but it would be interesting to compare of different models address these sorts of questions. And interesting to see how their internal models do as well.
For quite a while-- 50's, 60's, 70's - SF writers generally believed that the future would see an Earth in which national governments took back seats to the UN or other form of world government. I don't think that was narrative convenience, but what felt like reasonable extrapolation.
October 3, 2026 at 1:34 PM
For quite a while-- 50's, 60's, 70's - SF writers generally believed that the future would see an Earth in which national governments took back seats to the UN or other form of world government. I don't think that was narrative convenience, but what felt like reasonable extrapolation.
Parce qu'elle résulte à une extrapolation à l'ensemble des contribuables de l'ensemble des contrôles effectués, ce qui serait correct si les contrôles étaient menés de façon aléatoire.
Mais c'est loin d'être le cas, parce que les contrôles ciblent les contribuables et les dispositifs les plus
Mais c'est loin d'être le cas, parce que les contrôles ciblent les contribuables et les dispositifs les plus
October 3, 2026 at 11:48 AM
Parce qu'elle résulte à une extrapolation à l'ensemble des contribuables de l'ensemble des contrôles effectués, ce qui serait correct si les contrôles étaient menés de façon aléatoire.
Mais c'est loin d'être le cas, parce que les contrôles ciblent les contribuables et les dispositifs les plus
Mais c'est loin d'être le cas, parce que les contrôles ciblent les contribuables et les dispositifs les plus
It did start to rip apart — the rudder tore off.
I don’t know how much is data and how much is extrapolation.
I don’t know how much is data and how much is extrapolation.
October 3, 2026 at 10:12 AM
It did start to rip apart — the rudder tore off.
I don’t know how much is data and how much is extrapolation.
I don’t know how much is data and how much is extrapolation.
Recursive Reasoning or Statistical Extrapolation? In-Context Learning in Multi-Agent Interdependent Decision-Making: Guangnan Ye
NEP/RePEc link
to paper
d.repec.org
October 3, 2026 at 9:52 AM
Recursive Reasoning or Statistical Extrapolation? In-Context Learning in Multi-Agent Interdependent Decision-Making: Guangnan Ye
Pain is a *feeling.*
...My grounding is in biology. You have no way to transmit pain, you have no pain. I find their definition of "pain" as unbelievable.
*The models are meant to model human behavior.*
There's no added function, it's extrapolation from a mathematical model.
...My grounding is in biology. You have no way to transmit pain, you have no pain. I find their definition of "pain" as unbelievable.
*The models are meant to model human behavior.*
There's no added function, it's extrapolation from a mathematical model.
October 3, 2026 at 8:26 AM
Pain is a *feeling.*
...My grounding is in biology. You have no way to transmit pain, you have no pain. I find their definition of "pain" as unbelievable.
*The models are meant to model human behavior.*
There's no added function, it's extrapolation from a mathematical model.
...My grounding is in biology. You have no way to transmit pain, you have no pain. I find their definition of "pain" as unbelievable.
*The models are meant to model human behavior.*
There's no added function, it's extrapolation from a mathematical model.
オンポリシー蒸留（OPD）は、生徒モデルが自身の軌跡上で教師モデルの次トークン分布と一致するように学習させる手法であり、実証的に大きな成果を上げている。
一般化された変種により、学習者は出力空間における暗黙の報酬を推論することで、教師の能力を超えることができる。
しかし、言語モデルのヘッドはこの変化を異方的に減衰させる。つまり、教師の隠れ状態にエンコードされた変化の大部分は、その重みのごく一部としてロジットに伝達され、出力空間の外挿が依存するサンプリングされたトークンの対数確率比にはノイズが混入し、そのノイズが外挿によって増幅されることで、学習が不安定になる。
強化学習（RL）によ...
一般化された変種により、学習者は出力空間における暗黙の報酬を推論することで、教師の能力を超えることができる。
しかし、言語モデルのヘッドはこの変化を異方的に減衰させる。つまり、教師の隠れ状態にエンコードされた変化の大部分は、その重みのごく一部としてロジットに伝達され、出力空間の外挿が依存するサンプリングされたトークンの対数確率比にはノイズが混入し、そのノイズが外挿によって増幅されることで、学習が不安定になる。
強化学習（RL）によ...
October 3, 2026 at 12:05 AM
オンポリシー蒸留（OPD）は、生徒モデルが自身の軌跡上で教師モデルの次トークン分布と一致するように学習させる手法であり、実証的に大きな成果を上げている。
一般化された変種により、学習者は出力空間における暗黙の報酬を推論することで、教師の能力を超えることができる。
しかし、言語モデルのヘッドはこの変化を異方的に減衰させる。つまり、教師の隠れ状態にエンコードされた変化の大部分は、その重みのごく一部としてロジットに伝達され、出力空間の外挿が依存するサンプリングされたトークンの対数確率比にはノイズが混入し、そのノイズが外挿によって増幅されることで、学習が不安定になる。
強化学習（RL）によ...
一般化された変種により、学習者は出力空間における暗黙の報酬を推論することで、教師の能力を超えることができる。
しかし、言語モデルのヘッドはこの変化を異方的に減衰させる。つまり、教師の隠れ状態にエンコードされた変化の大部分は、その重みのごく一部としてロジットに伝達され、出力空間の外挿が依存するサンプリングされたトークンの対数確率比にはノイズが混入し、そのノイズが外挿によって増幅されることで、学習が不安定になる。
強化学習（RL）によ...
Uhhm... AI extrapolation software or some rubbish. You like?
October 2, 2026 at 11:01 PM
Uhhm... AI extrapolation software or some rubbish. You like?
The extrapolation for math with less than 100% accuracy makes perfect sense to me from the analogies I've used to understand what LLMs do.
My personal terminology for the weights was an n-dimensional calibration curve. Enough examples of arithmetic and you can make really good correlation output
My personal terminology for the weights was an n-dimensional calibration curve. Enough examples of arithmetic and you can make really good correlation output
October 2, 2026 at 10:36 PM
The extrapolation for math with less than 100% accuracy makes perfect sense to me from the analogies I've used to understand what LLMs do.
My personal terminology for the weights was an n-dimensional calibration curve. Enough examples of arithmetic and you can make really good correlation output
My personal terminology for the weights was an n-dimensional calibration curve. Enough examples of arithmetic and you can make really good correlation output
Fanboys. These people are very rich Culture fanboys who badly want Claude and its ilk to be proto-Minds.
But Iain Banks did that by authorial fiat. The Culture books aren't extrapolation, and longing is not going to make Claude grow up to be the Sleeper Service, or even Flere-Imsaho.
But Iain Banks did that by authorial fiat. The Culture books aren't extrapolation, and longing is not going to make Claude grow up to be the Sleeper Service, or even Flere-Imsaho.
It's hard to tell which is weirder: the fact that these guys seriously suggest that models like Claude have consciousness/sentience or that they actually invest time and money (weird!) trying to convince the Pope as well as theologians and legal scholars of this belief.
This is insane.
October 2, 2026 at 9:22 PM
Fanboys. These people are very rich Culture fanboys who badly want Claude and its ilk to be proto-Minds.
But Iain Banks did that by authorial fiat. The Culture books aren't extrapolation, and longing is not going to make Claude grow up to be the Sleeper Service, or even Flere-Imsaho.
But Iain Banks did that by authorial fiat. The Culture books aren't extrapolation, and longing is not going to make Claude grow up to be the Sleeper Service, or even Flere-Imsaho.
Yeah, I should clarify my meaning here. To me, "the tool" is the tech that allows us to use training data to twist vectors through hyperspace and end up with a preposterously good correlation and extrapolation engine. Getting the training data is part of the "how we use it" phase.
October 2, 2026 at 7:12 PM
Yeah, I should clarify my meaning here. To me, "the tool" is the tech that allows us to use training data to twist vectors through hyperspace and end up with a preposterously good correlation and extrapolation engine. Getting the training data is part of the "how we use it" phase.
Draft of a Real Cognitohazard Diamond
October 2, 2026 at 5:47 PM
Draft of a Real Cognitohazard Diamond
Reading comprehension & higher extrapolation continue to be important.
If you legit can't tell whether it's AI because the passage is short or a human writer just kinda sucks, you can always just...hit the back button? Put it down? Without evidence or confession, it's just a witch hunt. And we
If you legit can't tell whether it's AI because the passage is short or a human writer just kinda sucks, you can always just...hit the back button? Put it down? Without evidence or confession, it's just a witch hunt. And we
October 2, 2026 at 5:31 PM
Reading comprehension & higher extrapolation continue to be important.
If you legit can't tell whether it's AI because the passage is short or a human writer just kinda sucks, you can always just...hit the back button? Put it down? Without evidence or confession, it's just a witch hunt. And we
If you legit can't tell whether it's AI because the passage is short or a human writer just kinda sucks, you can always just...hit the back button? Put it down? Without evidence or confession, it's just a witch hunt. And we
AI data centers are holding pens for slaves, and as such are immoral and shouldn’t be built is an excellent take.
(Yes, I know thats wild extrapolation from the original post and article)
(Yes, I know thats wild extrapolation from the original post and article)
Daniel
@growing_daniel
My favorite part of this article is one of the visiting scholars pointing out that by Olah’s logic he is by far the worst slaver in history and he must immediately give up this abhorrent and immensely lucrative project, which anthropic people ignored.
@growing_daniel
My favorite part of this article is one of the visiting scholars pointing out that by Olah’s logic he is by far the worst slaver in history and he must immediately give up this abhorrent and immensely lucrative project, which anthropic people ignored.
October 2, 2026 at 5:28 PM
AI data centers are holding pens for slaves, and as such are immoral and shouldn’t be built is an excellent take.
(Yes, I know thats wild extrapolation from the original post and article)
(Yes, I know thats wild extrapolation from the original post and article)
This month with probably be adjusted also as businesses only had one day to compile hiring data and send to the gov't. The gov't always does an extrapolation to get to the final number since thousands of businesses won't have provided data.
October 2, 2026 at 3:37 PM
This month with probably be adjusted also as businesses only had one day to compile hiring data and send to the gov't. The gov't always does an extrapolation to get to the final number since thousands of businesses won't have provided data.
I suppose the idea is to avoid starting with that feeling and then going on to automatic chains of association and extrapolation; just to note the feeling as something in itself?
October 2, 2026 at 3:35 PM
I suppose the idea is to avoid starting with that feeling and then going on to automatic chains of association and extrapolation; just to note the feeling as something in itself?
by extrapolation ... S̲cience&̲M̲aths
October 2, 2026 at 3:24 PM
by extrapolation ... S̲cience&̲M̲aths
Sorry, how TF do you assign anyone who *can* stand up an "age of 100"? Most people are dead at 100. I'm no medic but I am a mathematician and there's some nasty extrapolation going on there.
October 2, 2026 at 3:19 PM
Sorry, how TF do you assign anyone who *can* stand up an "age of 100"? Most people are dead at 100. I'm no medic but I am a mathematician and there's some nasty extrapolation going on there.
“Women make up more than half the population, but most of what we know about women’s biology comes from extrapolation of studies in men.”
Contents | Science 394, 6819
www.science.org
October 2, 2026 at 3:18 PM
“Women make up more than half the population, but most of what we know about women’s biology comes from extrapolation of studies in men.”
Victoria Monet is a GENIUS for tucking a Voyage to Atlantis extrapolation in this record.
Come Back To You by Victoria Monét on Apple Music
Song · 2026 · Duration 2:10
music.apple.com
October 2, 2026 at 3:14 PM
Victoria Monet is a GENIUS for tucking a Voyage to Atlantis extrapolation in this record.
Petite extrapolation. Monsieur le Ministre n'a pas dit si c'était pas la police, les manifestants ou d'autres sources. Etrange !
October 2, 2026 at 1:49 PM
Petite extrapolation. Monsieur le Ministre n'a pas dit si c'était pas la police, les manifestants ou d'autres sources. Etrange !
that is an absurd, bad-faith extrapolation of what i said. come on now
October 2, 2026 at 12:18 PM
that is an absurd, bad-faith extrapolation of what i said. come on now
Now playing on MotoStorie: I J U S T W A N N A (ft. Griff [extrapolation]) by SO SHA! Tune in now: https://radiofonico.it #motostorie #radio
October 2, 2026 at 9:21 AM
Now playing on MotoStorie: I J U S T W A N N A (ft. Griff [extrapolation]) by SO SHA! Tune in now: https://radiofonico.it #motostorie #radio