Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,179 @@
|
|
| 1 |
---
|
| 2 |
license: mit
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
+
language:
|
| 4 |
+
- zh
|
| 5 |
+
- en
|
| 6 |
+
pretty_name: Wo Dynamics (WD)
|
| 7 |
+
size_categories:
|
| 8 |
+
- n<1K
|
| 9 |
+
tags:
|
| 10 |
+
- reasoning
|
| 11 |
+
- formal-reasoning
|
| 12 |
+
- theory
|
| 13 |
+
- epistemology
|
| 14 |
+
- complex-systems
|
| 15 |
+
- social-science
|
| 16 |
+
configs:
|
| 17 |
+
- config_name: concepts
|
| 18 |
+
data_files: data/concepts.jsonl
|
| 19 |
+
default: true
|
| 20 |
+
- config_name: papers
|
| 21 |
+
data_files: data/papers.jsonl
|
| 22 |
+
- config_name: formalizations
|
| 23 |
+
data_files: data/formalizations.jsonl
|
| 24 |
---
|
| 25 |
+
|
| 26 |
+
# 渦動力學 Wo Dynamics (WD)
|
| 27 |
+
|
| 28 |
+
一套公理化的理論框架:以同一組基本詞描述物理、生物、AI 與社會,並據此標出任何智能體在認識上的結構性極限。以明文定義、逐步形式化的推導,以及每一步推導與原文句子的對齊發布。
|
| 29 |
+
|
| 30 |
+
An axiomatic framework that describes physics, biology, AI and society with one set of foundational terms, and uses it to mark the structural limits of what any agent can know. Published as explicit definitions, step-by-step formalized derivations, and sentence-level alignment between every derivation step and the original text.
|
| 31 |
+
|
| 32 |
+
正本在 GitHub:[mthreetw/Wo_Dynamics](https://github.com/mthreetw/Wo_Dynamics)。本資料集在每次推到 `main` 時自動同步。
|
| 33 |
+
|
| 34 |
+
Source of truth: [mthreetw/Wo_Dynamics](https://github.com/mthreetw/Wo_Dynamics) on GitHub. This dataset is synced automatically on every push to `main`.
|
| 35 |
+
|
| 36 |
+
## 這套論文集要做什麼 / What this collection sets out to do
|
| 37 |
+
|
| 38 |
+
從一個最小的本體論出發:一個宇宙、一部只往後追加的歷史,加上依權重選出結果的坍塌。由此立下四條鐵律,再用同一套底層,一路往上描述能被歷史改變的個體、個體之間的互動、生產,以及「我們」、邊界、身分這類社會構件。
|
| 39 |
+
|
| 40 |
+
Starting from a minimal ontology — one universe, one history that only grows by appending, and collapses that select outcomes by weight — the framework lays down four iron laws, then builds upward on the same foundation: entities whose history shapes their next step, interaction between them, production, and social constructs such as "we", boundaries and identities.
|
| 41 |
+
|
| 42 |
+
目標有三個:
|
| 43 |
+
|
| 44 |
+
It has three aims:
|
| 45 |
+
|
| 46 |
+
1. **統一底層 / One foundation.**
|
| 47 |
+
物理、生物、AI、社會使用同一組基本詞來描述,不為每一層另立生成規則。
|
| 48 |
+
Physics, biology, AI and society are described with the same small set of terms; no separate generative rule is added for any layer.
|
| 49 |
+
2. **標出結構性的認識極限 / Structural limits of knowing.**
|
| 50 |
+
任何智能體,不論是人還是 AI,對他者、對自己、對未來,能知道什麼、不能知道什麼,以及為什麼這些限制不會隨工程進步而消失。
|
| 51 |
+
What any agent, human or AI, can and cannot know about others, about itself and about the future, and why these limits do not disappear with better engineering.
|
| 52 |
+
3. **作為處理高維資料的理論 / A theory for handling high-dimensional data.**
|
| 53 |
+
當研究對象是歷史依賴、不被鎖死的系統時,壓縮、預測、比對各自能保證什麼、不能保證什麼。應用範圍是生物學、AI 與社會科學。框架規定的是能主張什麼,不是處理資料的演算法。
|
| 54 |
+
When the object of study is a history-dependent system that is not locked in, what compression, prediction and comparison can and cannot guarantee. Intended fields of application: biology, AI and the social sciences. The framework specifies what can be claimed; it is not a data-processing algorithm.
|
| 55 |
+
|
| 56 |
+
### 論文樹的層次 / Layers of the paper tree
|
| 57 |
+
|
| 58 |
+
依引用關係排列,上游在前。 / Listed in dependency order, upstream first.
|
| 59 |
+
|
| 60 |
+
| 層 Layer | 論文 Papers | 回答的問題 Question addressed |
|
| 61 |
+
|---|---|---|
|
| 62 |
+
| 本體 Ontology | 暗宇宙的本體論 The Ontology of the Dark Universe;渦動力學 Wo Dynamics | 世界由什麼構成,什麼東西的歷史會影響它的下一步。What the world consists of, and what kind of entity has a next step shaped by its history. |
|
| 63 |
+
| 工具與極限 Tools and limits | 四條律相關工具及範例 Tools and Examples Related to the Four Laws;三重幻覺與被迫的賭注 Triple Illusion and the Forced Wager | 四條律能推出什麼;一個「我」為什麼看不全來處、自己與未來。What the four laws imply; why a self cannot fully see its origin, itself, or its future. |
|
| 64 |
+
| 互動 Interaction | 互動動力學-共同外生背景下的二階網路 Interaction Dynamics - Second-Order Networks under a Common Exogenous Background;互動動力學-陌生人迴路中的專屬歷史變動與更新 Interaction Dynamics - Changes and Updates of Dedicated History in Stranger Loops | 兩個以上的個體怎麼互相影響,文化與默契怎麼出現。How two or more entities affect each other; how culture and tacit coordination arise. |
|
| 65 |
+
| 生物人 Biological humans | 生產動力學 Production Dynamics;生物人個體輸出傾向動力學 Dynamics of Output Tendency in the Biological Human Individual;生物人輸出格局動力學 Dynamics of Output Patterns among Biological Humans | 個人與群體的行為怎麼記錄、歸納,哪些推不出。How individual and group behavior is recorded and generalized, and what cannot be inferred. |
|
| 66 |
+
| 構件 Constructs | 邊界的幻覺構件動力學 The Dynamics of Boundary Illusory Constructs;幻覺構件動力學 Dynamics of Illusory Constructs | 「我們」、邊界、概念這類東西怎麼從單純的物理反應生成。How "we", boundaries and concepts arise from mere physical responses. |
|
| 67 |
+
|
| 68 |
+
目前完成的是基礎層,應用層尚未開始。
|
| 69 |
+
|
| 70 |
+
The current papers form the foundation layer. The application layer has not started.
|
| 71 |
+
|
| 72 |
+
## 資料內容 / Contents
|
| 73 |
+
|
| 74 |
+
| 子集 Config | 一列是 One row is |
|
| 75 |
+
|---|---|
|
| 76 |
+
| `concepts`(預設 default) | 一個概念:中英文定義、public/private、定義依賴,以及提到它的形式化陳述。A concept: Chinese and English definitions, public/private status, definition dependencies, and the formalized statements that mention it. |
|
| 77 |
+
| `papers` | 一篇論文:中文全文與英文翻譯,以及概念表與形式化文件。A paper: the full Chinese text and English translation, plus its concept table and formalization file. |
|
| 78 |
+
| `formalizations` | 一篇論文的形式化:每條陳述一個節點,含角色(前提、推導、例……)、標籤、述詞關係、解析後的來源,以及引用的原文句碼。A paper's formalization: one node per statement, with role (premise, derivation, example, …), label, predicate relations, resolved sources, and the codes of the original sentences it cites. |
|
| 79 |
+
|
| 80 |
+
目前共 11 篇論文、約 290 個概念,形式化中約有 1,300 步推導。
|
| 81 |
+
|
| 82 |
+
The current release has 11 papers and about 290 concepts; the formalizations contain about 1,300 derivation steps.
|
| 83 |
+
|
| 84 |
+
`data/寫作用概念.md` 是全部 public 概念的精簡清單,寫新論文時交給模型當上下文。
|
| 85 |
+
|
| 86 |
+
`data/寫作用概念.md` is a compact list of all public concepts, meant to be given to a model as context when writing a new paper.
|
| 87 |
+
|
| 88 |
+
```python
|
| 89 |
+
from datasets import load_dataset
|
| 90 |
+
|
| 91 |
+
concepts = load_dataset("mthreetw/Wo-Dynamics-WD", "concepts", split="train")
|
| 92 |
+
papers = load_dataset("mthreetw/Wo-Dynamics-WD", "papers", split="train")
|
| 93 |
+
formalizations = load_dataset("mthreetw/Wo-Dynamics-WD", "formalizations", split="train")
|
| 94 |
+
```
|
| 95 |
+
|
| 96 |
+
### 識別與版本 / Identifiers and versions
|
| 97 |
+
|
| 98 |
+
- 概念的身分是「論文 UUID + 概念 ID」,例如 `ed71bef7-0f22-4443-bc1d-4d1ecfe81967.c12`;顯示名稱不屬於身分。
|
| 99 |
+
A concept is identified by `paper UUID + concept ID`, for example `ed71bef7-0f22-4443-bc1d-4d1ecfe81967.c12`. Display names are not part of the identity.
|
| 100 |
+
- 論文定稿(commit)後 ID 凍結;廢棄的條目保留並標記,不刪除,ID 永不重用。跨版本的引用因此穩定。
|
| 101 |
+
Once a paper is committed, its concept IDs are frozen. Deprecated concepts are kept and marked, never deleted, and their IDs are never reused, so references across versions stay stable.
|
| 102 |
+
- 中文論文的每個句子有依內容算出的十位句碼;形式化陳述以句碼引用原文。句子一改,句碼就變。
|
| 103 |
+
Each sentence of a Chinese paper has a 10-digit code computed from its content. Formalized statements cite these codes; if a sentence changes, its code changes.
|
| 104 |
+
- `formalizations` 帶有 `schema_version` 欄位(目前為 `1.2`)。
|
| 105 |
+
`formalizations` carries a `schema_version` field (currently `1.2`).
|
| 106 |
+
|
| 107 |
+
## 這份資料的不同之處 / What makes this dataset different
|
| 108 |
+
|
| 109 |
+
1. **每一步推導都列出來源,並對回原句。** 來源只能是前提、前面的步驟、概念的定義或鐵律,並解析成概念 ID。
|
| 110 |
+
**Every derivation step lists its sources and points back to the original sentences.** Sources are restricted to premises, earlier steps, concept definitions, or laws, and are resolved to concept IDs.
|
| 111 |
+
2. **概念是穩定、有版本的介面。** 下游論文只能引用上游論文的 `public` 概念。
|
| 112 |
+
**Concepts are stable, versioned interfaces.** Downstream papers may only cite `public` concepts of upstream papers.
|
| 113 |
+
3. **每篇論文在 commit 前都通過驗收流程:** 工具檢查(連結、格式、與上一次 commit 比對),再經多個獨立 AI 稽核與 AI 裁判。語意上的正確性**沒有**機器證明(`checks.semantic_proof = not_checked`)。
|
| 114 |
+
**Each paper passes a validation pipeline before it is committed:** a validation tool (link integrity, format, comparison against the last commit), followed by independent multi-AI audits and an AI judge. Semantic validity is **not** machine-proven (`checks.semantic_proof = not_checked`).
|
| 115 |
+
|
| 116 |
+
### 範例 / Example
|
| 117 |
+
|
| 118 |
+
《幻覺構件動力學》中的一步推導(`formalizations`,標籤 `結論三之三`):
|
| 119 |
+
|
| 120 |
+
A derivation step from *Dynamics of Illusory Constructs* (`formalizations`, label `結論三之三`):
|
| 121 |
+
|
| 122 |
+
- **陳述 Statement**:五項工程能力都是衡量載體某類動作或保留的指標,不是依次執行的工序;同一機制可以同時落在幾項指標所衡量的面向上。
|
| 123 |
+
The five engineering capacities are measures of what a carrier can do or retain, not stages executed in sequence; one mechanism can fall under several of them at once.
|
| 124 |
+
- **來源 Sources**:`c12`(工程能力)與 `c13`–`c17`(容納、運算、記憶、壓縮、構造能力)的定義。
|
| 125 |
+
The definitions of `c12` (engineering capacity) and `c13`–`c17` (containment, operation, memory, compression, structure capacity).
|
| 126 |
+
- **引用的原句 Cited sentences**(`body_citations`):
|
| 127 |
+
- 每一項都是指標,衡量載體作某類動作或某類保留能到什麼程度……
|
| 128 |
+
- 五項不是依次執行的工序,不是線性串接。
|
| 129 |
+
- 同一機制可以同時落在幾項指標所衡量的面向上,例如帶參數的壓縮、受記憶影響的注意力分配。
|
| 130 |
+
|
| 131 |
+
## 建議用途 / Suggested uses
|
| 132 |
+
|
| 133 |
+
目前還沒有官方基準與分數。以下用法的答案都能由資料本身判定,使用者不必先接受這套理論。
|
| 134 |
+
|
| 135 |
+
There is no official benchmark or baseline yet. The following uses can all be graded from the data itself, without accepting the theory.
|
| 136 |
+
|
| 137 |
+
- **來源是否充分 / Premise sufficiency**:給結論與所列來源,判斷來源夠不夠推出結論;把正確的步驟改壞,就能產生反例。
|
| 138 |
+
Given a conclusion and its listed sources, decide whether the sources are enough. Negative examples can be generated by corrupting correct steps.
|
| 139 |
+
- **只依給定定義推理 / Definition-faithful reasoning**:不少概念的定義和日常用法不同。把概念清單當上下文給模型,測它是照給定定義推理,還是退回自己原有的知識。
|
| 140 |
+
Many concepts are defined differently from everyday usage. Given the concept list as context, test whether a model reasons from the given definitions or falls back on its prior knowledge.
|
| 141 |
+
- **依賴追蹤 / Dependency tracking**:某條前提被否定時,哪些結論失去來源?答案可由來源鏈機械算出。
|
| 142 |
+
If a premise is rejected, which conclusions lose their source? The answer can be computed mechanically from the source chains.
|
| 143 |
+
- **審查主張 / Claim auditing**:檢查生物學、AI 或社會科學論文的實證主張,有沒有超出它的資料能支持的範圍。
|
| 144 |
+
Check whether empirical claims in biology, AI or social science papers go beyond what their data can support.
|
| 145 |
+
|
| 146 |
+
## 給機器學習讀者的對照 / Reading guide for ML readers
|
| 147 |
+
|
| 148 |
+
以下是大致的對應,**只供閱讀時定位**,不是定義;每個詞的意思一律以概念表為準。
|
| 149 |
+
|
| 150 |
+
Rough correspondences, **for orientation only**. They are not definitions; the meaning of every term is fixed by its concept table entry.
|
| 151 |
+
|
| 152 |
+
| 框架中 In this framework | 大致對應 Roughly corresponds to |
|
| 153 |
+
|---|---|
|
| 154 |
+
| 前窗/後窗 Earlier window / later window | 訓練資料/測試資料 Training data / test data |
|
| 155 |
+
| 案例認定 Case attribution | 樣本身分、資料串接 Sample identity, record linkage |
|
| 156 |
+
| 允許誤差 Allowed error | 評估指標與門檻 Evaluation metric and threshold |
|
| 157 |
+
| 在已登記的案例中可交換 Commutative over registered cases | 在已測資料上表現好 Good performance on the data tested so far |
|
| 158 |
+
| 在範圍內可交換是一項主張 Commutative within a range is a claim | 分布外的泛化無法由測試確認 Out-of-distribution generalization cannot be confirmed by testing |
|
| 159 |
+
| 運算能力再高,也補不回容納時丟掉的差異 Differences lost at containment cannot be recovered by more computation | 資料處理不等式 Data processing inequality |
|
| 160 |
+
| 不被鎖死的對象,單次結果不保證可預測 Single outcomes of a non-locked-in system are not guaranteed predictable | 不可約的雜訊、貝氏誤差 Irreducible noise, Bayes error |
|
| 161 |
+
| 份量不能相加 Impact is not additive | 交互作用、歷史依賴 Interaction effects, history dependence |
|
| 162 |
+
|
| 163 |
+
## 限制 / Limitations
|
| 164 |
+
|
| 165 |
+
- **這不是實證資料**,也不是已驗證的科學理論,而是一組定義與推導。
|
| 166 |
+
**Not empirical data**, and not a validated scientific theory. It is a set of definitions and derivations.
|
| 167 |
+
- **自洽性由 AI 稽核判定。** 不同的 AI 模型可能有相同的盲區。
|
| 168 |
+
**Internal consistency is judged by AI audit.** Different AI models may share the same blind spots.
|
| 169 |
+
- **目前只有基礎層。** 應用層的論文尚未發布。
|
| 170 |
+
**Foundation layer only.** Application papers are not yet released.
|
| 171 |
+
- **概念名稱不帶日常意思。** 特別是「幻覺構件」(*Illusory Construct*) 與機器學習的「幻覺」(hallucination) 無關;「死亡」「權力」「我」等也都是框架內的定義。
|
| 172 |
+
**Terms do not carry their everyday meaning.** In particular, *Illusory Construct* (幻覺構件) has nothing to do with "hallucination" in machine learning; *Death*, *Power*, *Self* and many others are defined within the framework.
|
| 173 |
+
|
| 174 |
+
## 授權與聯絡 / License and contact
|
| 175 |
+
|
| 176 |
+
MIT。作者:黃正宇 / Cheng Yu Huang(mthree.tw@gmail.com)。
|
| 177 |
+
|
| 178 |
+
MIT License. Author: 黃正宇 / Cheng Yu Huang (mthree.tw@gmail.com).
|
| 179 |
+
|