エルパト

Adrian Moore on Kant's "Epistemic Spectacles" Kant agreed with the empiricists that all knowledge must relate to experience but he didn't think knowledge was simply read off from experience. Moore explains: "When we derived knowledge of the world through the deliverances of experience, it was because we ourselves were appropriately receptive to what was out there. And being appropriately receptive meant having certain faculties in terms of which we interpreted what was out there." In other words, the mind isn't a blank slate passively receiving the world. It actively shapes what we perceive. "It's as if we were operating with a pair of spectacles, epistemological spectacles through which we interpret what the world throws at us." We don't see raw reality. We see reality filtered through the structures our minds bring to it categories, forms of intuition, frameworks of understanding that we can't simply take off. This is Kant's great reversal of the philosophical tradition before him. Rather than asking "does the mind conform to the world?", Kant asked: "does the world conform to the mind?" The answer, for Kant, was yes because experience itself is only possible through the lenses we already wear. What does this mean practically? It means that whenever you think you're simply "observing" something neutrally, you're already interpreting it. Your frameworks, assumptions, and prior categories are always already in play.

The role of memorization and knowledge is to cache & reuse past cognitive work. It should be leveraged as a way to speed up cognition, not as a *replacement* for cognition.

これめちゃくちゃ面白いので、引用元とその反論も含めて読んでみて下さい。 すごく適当に言えば、物事の原因を最小単位まで突き詰めたら、因果が自動的に分かる？という問いを批判して疑問を投げつけてる。

めっちゃ面白かったので自分なりの内容紹介。著者はAnthropic研究者。 ニューラルネットは簡単な特徴と難しい特徴を両方完璧に学習しても、内部表現は簡単な方に極端に偏る。 PCAなどの標準的な分析手法は、この偏りのせいで重要な計算を見落とす。 強いシグナル＝重要という前提を疑えという話。

A common experience: Many brilliant scientists cannot grasp elementary philosophical distinctions. Last night I could not get a colleague to understand the difference between the "hard" (sentience, subjectivity, experience) and "easy" (reportability, information access) senses of "consciousness." Nor the difference between a definition of consciousness and various explanations of consciousness. Hypothesis: scientists tend to equate rigorous thinking with mechanistic explanation, and don't recognize that abstract concepts requires sharp analysis as well.

This tweet seems to have gotten some spirited pushback, so I think it's important to clarify what I think Kauffman is saying here: starting from string theory or quarks, you could potentially derive human hearts as one of multiple events/artifacts in the universe of possibilities. But the *causal* relationship between the structure of the heart (its ability to pump blood) and its fundamental role in keeping organisms alive and therefore making the biosphere possible is an evolutionary accident. Physics *by itself* cannot capture this contingency. Perhaps we need to indicate the difference by saying "reduced to" vs "explained by".

情報圧縮について話してみます。 チャートが全ての情報である世界にいると仮定します。 このAAVEのチャートの情報は、ただの時間毎の数字の塊です。 情報を生のまま受け取れば、全ての数字をそのまま脳内に入れるのと同じです。 でも人間はそんなことしないですよね。情報を圧縮して、構造情報を持つ

Engineering PhDs (D.Eng) in China are getting wholesaled out now. Company employees take three months of classes per year and walk away with the degree straight from national policy. The state no longer trusts universities to drive real tech progress. Tang boss built Zhipu AI and still didnt get academician status. They think enterprises do it better while unis just pump out low quality papers. Unis are told to focus on talent cultivation and hand R&D to companies so freshmen might as well go straight to internships to learn by doing. Yet good company labs still pump out papers and undergrads get thrown into research groups. Young profs chase papers and titles old ones chase grants and cash. Nobody gets judged on actual teaching. Undergrad research only exists because everyone knows these profs wont teach so students chase the cutting edge themselves. Leadership wisdom at its finest write piles of equations no reviewer understands and keep managing upwards.

他の事せずに真面目に表現学習の時間の座標漂流対策してる。 ①変数追加→完璧に時系列で別れて同定できない ②設計完璧にする→勾配計算に300時間かかる😇 ③今までの設計でも成果あったから計算続けたらいける→特徴ある漂流しないクラスタの出現頻度低い ②を考えるのに8時間かけた。いらついた😇

培养research taste，就是培养自己的品菜能力。 但品菜这件事，比做菜更难教。 做菜可以给你食谱，给你计时器，给你步骤。品菜没有标准答案——你只能说"这道菜好"，说不清楚为什么好。但真正会品的人，一口下去就知道：咸了、火候差了半分钟、葱放早了。 research taste也是这样。你能感觉到一篇paper有问题，但说不出来。你看到一个idea，直觉告诉你这个方向不对，但你说不出理由。这种感觉，就是taste在发育的信号。 品菜能力是怎么来的？ > 吃得足够多。 没吃过一百道菜，不知道什么是"正常水平"。读paper也一样。taste的底层是distribution，读过足够多，才知道什么是平庸，什么是真的有东西。只读自己方向的paper，上来一道川菜以为是黑暗料理。 > 吃过真正好的。 光吃多不够。得吃过几道真正的好菜，才有锚点。去顶会找那几篇让所有人都在讨论的paper，认真读，想清楚它好在哪——不是"结果好"，而是它发现了什么别人没发现的东西。 > 自己做过，才真的会品。 没下过厨的人吃饭，只会说好吃不好吃。下过厨的人吃饭，吃的是决策。这个酱加多了，那一步火太急了。做过research，读paper才能读出作者当时的处境——这里为什么这么选，这里有没有埋坑，这里其实可以更简洁。 > 学会区分：食材的问题，还是厨艺的问题 (Topic不好还是paper execution不好)。 食材差，再好的厨师也难救。食材好，手艺差，暴殄天物。 > 品出"缺了什么"，比品出"哪里不好"更难。 普通的品菜能说出"这道菜太咸"。更高的品菜能力是："这道菜差一点, 如果有点酸就能把咸托出来"。 这就需要我们对"哪里不对"有更精准的理解力, 逼着自己问"然后呢"。 品菜能力最终是一种审美。审美不是天生的，但也很难靠刷题训练出来。它是在大量输入、少量输出、反复对照之后，慢慢沉淀下来的判断力。不需要每道菜都做得好，但得清楚地知道：什么是好菜，好在哪，差在哪，自己现在在哪个段位。 品菜能力，就是让你在厨房里不慌的底气。

（これは私の思考能力の欠落のせいでもありますが）今もそういう作業を結局しているのですが・・・、私はテクスト同士の「差」を取る作業をしているのが好きみたいでして——ですので、「注釈」とか、桑原先生的にいうと「◯◯◯を読むラカン」とかに関心があります——、結局それって複数のテクストにかんして四則演算的な計算をしているだけというか（残念ながら微積分まではいっていない気がします・・・）。そういう点で、そもそも人力で計算していただけなので、算盤しかなかったのが計算機になったら普通に乗り換えるように、人文業界内的におかしいかもしれませんがAI的なものに乗り換えることにはあまり違和感はないところがあります。むしろ、助かるくらいに思っているところがあります。