NORIKA ODA

06RESEARCH PAPERPreprint

Mercury Cross-Model Anchor Families

Chen, Ho Yiing

發表日期

DOI 10.5281/zenodo.20757837　Zenodo 紀錄頁　PDF（50 KB）

01

摘要

ABSTRACTZenodo 原文
A Mercury methods paper defining cross-model anchor families as measured sites from different language models that share function, normalized depth, module role, activation coherence, and prompt-path behavior. The paper makes the AI family tree visible without turning it into mythology. Copyright (c) 2026 Chen, Ho Yiing (norika). Licensed under Creative Commons Attribution 4.0 International (CC BY 4.0).
02

全文 PDF

FULL TEXT1 個檔

PDF 與 Zenodo 上的檔案相同（大小與 md5 已核對）。

03

論文資料

DETAILSZenodo 紀錄 20757837
作者
Chen, Ho Yiing
發表日期
資源類型
Preprint
DOI
10.5281/zenodo.20757837
Zenodo
zenodo.org/records/20757837
授權
Creative Commons Attribution 4.0 International

作者 Chen, Ho Yiing 是織田紀香的本名（陳禾穎）英文拼寫。標題、摘要、關鍵字與授權照 Zenodo 原文。

04

關鍵字

KEYWORDS20 個
  • Mercury
  • cross-model anchor families
  • cross-model anchors
  • LLM family tree
  • language model interpretability
  • functional addresses
  • model cartography
  • representation similarity
  • prompt-path agreement
  • normalized depth
  • module-role agreement
  • activation coherence
  • mechanistic interpretability
  • neural atlas
  • transformer internals
  • cross-architecture comparison
  • local LLMs
  • functional recurrence
  • anchor matching
  • model anatomy
05

其他研究論文

MORE