Tag: interpretability

29 items with this tag.

Jul 20, 2026
LLM 机制可解释性：一条方法谱系
- interpretability
Jul 20, 2026
读残差流：Logit Lens 到 Patchscopes
- interpretability
Jul 20, 2026
2025 前沿：谱系的收敛与对地基的反思
- interpretability
Jul 20, 2026
特征与叠加：字典学习这条线
- interpretability
Jul 20, 2026
干预与电路：从 Causal Tracing 到自动电路发现
- interpretability
Jul 20, 2026
LLM 内部坐标的排列不变性
Jul 20, 2026
Superposition（叠加）
- interpretability
Jul 20, 2026
Residual Stream（残差流）
- interpretability
Jul 20, 2026
Post-hoc vs Intrinsic 可解释性
- interpretability
Jul 20, 2026
Polysemanticity（多义性）
- interpretability
Jul 20, 2026
Linear Representation Hypothesis（线性表示假说）
- interpretability
Jul 20, 2026
Induction Head（归纳头）
- interpretability
Jul 20, 2026
Features 与 Dictionary Learning（特征与字典学习）
- interpretability
Jul 20, 2026
Faithfulness（忠实性）
- interpretability
Jul 20, 2026
Circuits（电路）
- interpretability
Jul 20, 2026
Causal Mediation（因果中介分析）
- interpretability
Jul 20, 2026
Activation Patching（激活patching / Causal Tracing）
- interpretability
Mar 27, 2025
Circuit Tracing / On the Biology of a Large Language Model
- interpretability
- LLM
Jan 27, 2025
参数空间可解释性：APD 与 SPD
- interpretability
- LLM
Jan 11, 2024
Patchscopes: A Unifying Framework for Inspecting Hidden Representations of LMs
- interpretability
- LLM
Oct 04, 2023
Towards Monosemanticity：用字典学习把语言模型解耦成单义特征
- interpretability
- LLM
Oct 02, 2023
Representation Engineering: A Top-Down Approach to AI Transparency
- interpretability
- LLM
Apr 28, 2023
ACDC：迈向机制可解释性的自动化电路发现
- interpretability
- LLM
Mar 05, 2023
DAS：在可解释因果变量与分布式神经表征之间寻找对齐
- interpretability
- LLM
Nov 01, 2022
Interpretability in the Wild：GPT-2 small 的 IOI 电路
- interpretability
- LLM
Sep 14, 2022
Toy Models of Superposition
- interpretability
- LLM
Mar 08, 2022
In-context Learning and Induction Heads
- interpretability
- LLM
Feb 10, 2022
Locating and Editing Factual Associations in GPT (ROME)
- interpretability
- LLM
Aug 31, 2020
interpreting GPT: the logit lens
- interpretability
- LLM

if_Guo

Explorer

Tag: interpretability

LLM 机制可解释性：一条方法谱系

读残差流：Logit Lens 到 Patchscopes

2025 前沿：谱系的收敛与对地基的反思

特征与叠加：字典学习这条线

干预与电路：从 Causal Tracing 到自动电路发现

LLM 内部坐标的排列不变性

Superposition（叠加）

Residual Stream（残差流）

Post-hoc vs Intrinsic 可解释性

Polysemanticity（多义性）

Linear Representation Hypothesis（线性表示假说）

Induction Head（归纳头）

Features 与 Dictionary Learning（特征与字典学习）

Faithfulness（忠实性）

Circuits（电路）

Causal Mediation（因果中介分析）

Activation Patching（激活patching / Causal Tracing）

Circuit Tracing / On the Biology of a Large Language Model

参数空间可解释性：APD 与 SPD

Patchscopes: A Unifying Framework for Inspecting Hidden Representations of LMs

Towards Monosemanticity：用字典学习把语言模型解耦成单义特征

Representation Engineering: A Top-Down Approach to AI Transparency

ACDC：迈向机制可解释性的自动化电路发现

DAS：在可解释因果变量与分布式神经表征之间寻找对齐

Interpretability in the Wild：GPT-2 small 的 IOI 电路

Toy Models of Superposition

In-context Learning and Induction Heads

Locating and Editing Factual Associations in GPT (ROME)

interpreting GPT: the logit lens