The paper investigates the reliability of attention‑head ablation as a causal inference tool in language models. Using GPT‑2 small, the authors find that a natural post‑projection zeroing method is almost uncorrelated with a corrected pre‑projection ablation and yields a completely different set of top‑5 important heads. They also show that binary accuracy can mask effects near performance floors or ceilings, whereas gold‑token log‑probability provides a graded signal. By employing a discovery/held‑out split and 1,000 matched random‑head and layer‑matched‑head controls, the corrected per‑head effect ranking remains highly stable (Spearman ρ = 0.974) and the top‑5 heads significantly outperform both control distributions (Monte Carlo p = 0.001). However, evidence for task specificity is weak on GPT‑2, and replication on DistilGPT‑2 confirms the intervention‑semantic and matched‑control findings.
arXiv:2606. 05378v1 Announce Type: new Abstract: We test whether a single screen-and-ablate recipe -- identify attention-head circuits by task-pattern selectivity, then verify by causal ablation against a matched-random null -- produces consistent mechanistic claims across model families.
arXiv:2607. 21692v1 Announce Type: new Abstract: Sparse attention reduces the cost of long contexts by allowing each query to read only selected parts of the input.
arXiv:2606. 08292v2 Announce Type: replace Abstract: Mechanistic studies often assign a component a role when removing it damages a behavior, its activation linearly encodes task information, and restoring that activation repairs the damage.
arXiv:2606. 08292v1 Announce Type: new Abstract: In mechanistic interpretability, attention heads are commonly elevated to role claims (e.
arXiv:2607. 20524v1 Announce Type: new Abstract: Mean cross-positional attention degradation is widely reported in transformer interpretability, yet whether it causally limits contextual retrieval remains untested.
arXiv:2607. 01002v1 Announce Type: cross Abstract: In long-context use, large language models frequently synthesize answers from the meaning of a relevant context span rather than literally copy-pasting them.
The paper investigates why vision‑language models like LLaVA‑1.5‑7B hallucinate objects in captions and proposes a targeted fix. By ranking attention heads whose image attention drops around hallucinated words, the authors identify 32 key heads and apply a head‑sliced LoRA adapter plus an inference‑time grounding controller. On COCO images, this combined method reduces hallucinated captions from 37% to 23% and hallucinated object mentions from 15.6% to 9.6%, while also lowering object recall.
arXiv:2606. 09876v1 Announce Type: new Abstract: Large language models often express high confidence in answers that are wrong.
The study investigates whether attention heads in large language models that align with human EEG signals are causally involved in model computation. By ablating these brain‑aligned heads during a pattern‑completion task, the authors find that while such heads contribute to performance, their removal is less disruptive than removing heads selected by attribution patching. The research also distinguishes two families of brain‑aligned heads—novelty and repetition heads—highlighting that novelty heads track human attention but are less critical than random ablation, whereas repetition heads modestly aid performance and align with abstract‑pattern representations.
arXiv:2609.38810v1 Announce Type: cross Abstract: As vision language models are increasingly deployed in clinical diagnosis, understanding how they internally resolve competing visual and textual sig...
arXiv:2609.39971v1 Announce Type: cross Abstract: Vision-language-action (VLA) models can exceed 90% success on in-distribution tasks and withstand nuisance changes that preserve the required action,...
As vision language models are increasingly deployed in clinical diagnosis, understanding how they internally resolve competing visual and textual signals becomes a safety imperative. Existing mechanis...