JoMA: Demystifying Multilayer Transformers via JOint Dynamics of MLP and Attention
Paper
•
2310.00535
•
Published
•
2
Papers - Observability and Interpretability
-
-
Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small
Paper • 2211.00593 • Published • 2 -
Rethinking Interpretability in the Era of Large Language Models
Paper • 2402.01761 • Published • 18 -
Does Circuit Analysis Interpretability Scale? Evidence from Multiple Choice Capabilities in Chinchilla
Paper • 2307.09458 • Published • 9 -
Sparse Autoencoders Find Highly Interpretable Features in Language Models
Paper • 2309.08600 • Published • 11 -
Interpretability at Scale: Identifying Causal Mechanisms in Alpaca
Paper • 2305.08809 • Published • 2 -
Natural Language Decomposition and Interpretation of Complex Utterances
Paper • 2305.08677 • Published • 2 -
Information Flow Routes: Automatically Interpreting Language Models at Scale
Paper • 2403.00824 • Published • 3 -
Structural Similarities Between Language Models and Neural Response Measurements
Paper • 2306.01930 • Published • 2 -
The Impact of Depth and Width on Transformer Language Model Generalization
Paper • 2310.19956 • Published • 9 -
The Data Provenance Initiative: A Large Scale Audit of Dataset Licensing & Attribution in AI
Paper • 2310.16787 • Published • 3 -
A Pretrainer's Guide to Training Data: Measuring the Effects of Data Age, Domain Coverage, Quality, & Toxicity
Paper • 2305.13169 • Published • 2 -
A Watermark for Large Language Models
Paper • 2301.10226 • Published • 8 -
Universal and Transferable Adversarial Attacks on Aligned Language Models
Paper • 2307.15043 • Published • 2 -
Vision Transformers Need Registers
Paper • 2309.16588 • Published • 69 -
Can Sensitive Information Be Deleted From LLMs? Objectives for Defending Against Extraction Attacks
Paper • 2309.17410 • Published • 4 -
On the Origin of LLMs: An Evolutionary Tree and Graph for 15,821 Large Language Models
Paper • 2307.09793 • Published • 45 -
Tools for Verifying Neural Models' Training Data
Paper • 2307.00682 • Published • 2 -
Shadow Alignment: The Ease of Subverting Safely-Aligned Language Models
Paper • 2310.02949 • Published • 2 -
Chain-of-Thought Reasoning Without Prompting
Paper • 2402.10200 • Published • 90 -
Building and Interpreting Deep Similarity Models
Paper • 2003.05431 • Published • 2 -
Long-form factuality in large language models
Paper • 2403.18802 • Published • 23 -
Unsolvable Problem Detection: Evaluating Trustworthiness of Vision Language Models
Paper • 2403.20331 • Published • 14 -
Locating and Editing Factual Associations in Mamba
Paper • 2404.03646 • Published • 3 -
BERT Rediscovers the Classical NLP Pipeline
Paper • 1905.05950 • Published • 1 -
Prompt-to-Prompt Image Editing with Cross Attention Control
Paper • 2208.01626 • Published • 1 -
LVLM-Intrepret: An Interpretability Tool for Large Vision-Language Models
Paper • 2404.03118 • Published • 17