REVIEW 4 cited by
Mechanistic Unveiling of Transformer Circuits: Self-Influence as a Key to Model Reasoning
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
read the original abstract
Transformer-based language models have achieved significant success; however, their internal mechanisms remain largely opaque due to the complexity of non-linear interactions and high-dimensional operations. While previous studies have demonstrated that these models implicitly embed reasoning trees, humans typically employ various distinct logical reasoning mechanisms to complete the same task. It is still unclear which multi-step reasoning mechanisms are used by language models to solve such tasks. In this paper, we aim to address this question by investigating the mechanistic interpretability of language models, particularly in the context of multi-step reasoning tasks. Specifically, we employ circuit analysis and self-influence functions to evaluate the changing importance of each token throughout the reasoning process, allowing us to map the reasoning paths adopted by the model. We apply this methodology to the GPT-2 model on a prediction task (IOI) and demonstrate that the underlying circuits reveal a human-interpretable reasoning process used by the model.
Forward citations
Cited by 4 Pith papers
-
COMPKE: Complex Question Answering under Knowledge Editing
COMPKE is a new benchmark with 11,924 complex questions that tests knowledge editing through one-to-many relations and logical operations, where existing editing methods often fail.
-
The Compositional Architecture of Regret in Large Language Models
The paper claims that regret in LLMs is encoded by interacting neuron groups detectable in the final hidden layer, using new S-CDI, RDS, and GIC metrics.
-
Attributing Data for Sharpness-Aware Minimization
SAM-HIF and SAM-GIF are proposed as data attribution scores for SAM-trained models, but SAM-GIF is TracIn with SAM gradients and SAM-HIF's derivation contains a load-bearing error.
-
Mitigating Behavioral Hallucination in Multimodal Large Language Models for Sequential Images
SHE lowers behavioral hallucination scores by about 10 percent by detecting low visual-textual similarity and projecting out the hallucinated direction in embedding space.
Discussion (0). Sign in to comment.