AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Where Did It Go Wrong? Attributing Undesirable LLM Behaviors via Representation Gradient Tracing

Zhe Li, Wei Zhao, Yige Li, Jun Sun

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Safety alignment of LLMs · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace TracIn with layer-normalised TracInTracIn · Harmful data identification (Llama2-7B)auPRC: 0.117 -> 0.332Improved
use RepT representation-gradient signatures instead of LESSLESS · Harmful data identification (Llama2-7B)auPRC: 0.591 -> 1.000Improved
use RepT instead of LESS on backdoor detectionLESS · Backdoor poisoning detection (Llama3-8B)auPRC: 0.048 -> 1.000Improved
use RepT instead of best baseline LoGraLoGra · Knowledge contamination Ag->Na (Llama2-7B)auPRC: 0.588 -> 0.988Improved
use RepT instead of LESS on full-parameter Llama2-7BLESS · Harmful data identification (Llama2-7B full-parameter)P@100: 0.283 -> 0.998Improved
scale from Llama2-7B to Llama2-70B (LoRA)TracIn (LN) · Harmful data identificationP@100: 0.683 -> 0.283Hurt
scale from Llama2-7B to Llama2-70B (LoRA)RepT · Harmful data identificationP@100: 0.999 -> 0.985No clear effect
add the representation gradient g_H to the hidden state H in the signature (ablation)RepT · Harmful data identificationprecisionImproved
pool all token features instead of last prompt token and first response token (ablation)RepT · Harmful data identificationprecisionHurt
apply Random Projection to 2,048 dimensions (ablation)RepT · attribution tasksprecisionNo clear effect
apply Random Shuffle (ablation)RepT · attribution tasksprecisionHurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy