REVIEW 5 cited by
Learning to Attribute with Attention
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Learning to Attribute with Attention
read the original abstract
Given a sequence of tokens generated by a language model, we may want to identify the preceding tokens that influence the model to generate this sequence. Performing such token attribution is expensive; a common approach is to ablate preceding tokens and directly measure their effects. To reduce the cost of token attribution, we revisit attention weights as a heuristic for how a language model uses previous tokens. Naive approaches to attribute model behavior with attention (e.g., averaging attention weights across attention heads to estimate a token's influence) have been found to be unreliable. To attain faithful attributions, we propose treating the attention weights of different attention heads as features. This way, we can learn how to effectively leverage attention weights for attribution (using signal from ablations). Our resulting method, Attribution with Attention (AT2), reliably performs on par with approaches that involve many ablations, while being significantly more efficient. To showcase the utility of AT2, we use it to prune less important parts of a provided context in a question answering setting, improving answer quality. We provide code for AT2 at https://github.com/MadryLab/AT2 .
Forward citations
Cited by 5 Pith papers
-
How Context Attribution Handles What the Model Already Knows
Context attribution methods cannot disentangle in-context from in-weight knowledge and assign unfaithful scores under overlap; new metrics and WMDP-Cyber++ quantify the failure.
-
Tokengeist: Multi-Turn Attribution Tracing in Agentic Conversations
Recursively attributing each conversation turn to earlier turns recovers multi-hop provenance paths (up to ~90% source recall) where single-pass attribution collapses to under 20%.
-
FlashRT: Towards Computationally and Memory Efficient Red-Teaming for Prompt Injection and Knowledge Corruption
FlashRT delivers 2x-7x speedup and 2x-4x GPU memory reduction for prompt injection and knowledge corruption attacks on long-context LLMs versus nanoGCG.
-
AgentWatcher: A Rule-based Prompt Injection Monitor
An attention-attribution plus rule-reasoning detector reduces prompt-injection success to near zero on agent and long-context benchmarks with small utility loss.
-
MaxShapley: Towards Incentive-compatible Generative Search with Fair Context Attribution
MaxShapley computes fair document attributions in generative QA by reducing Shapley value calculation to polynomial time via a max-sum utility, matching exact Shapley quality on HotPotQA, MuSiQUE, and MS MARCO while u...
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.