SPEX and ProxySPEX: Scalable Interaction Discovery for LLM Interpretability
Researchers from BAIR introduce SPEX (Spectral Explainer) and ProxySPEX, algorithms for identifying influential feature, data, and model-component interactions in LLMs at scale. The approach exploits sparsity, low-degreeness, and hierarchy properties to reframe interaction discovery as a sparse recovery problem using tools from signal processing and coding theory. ProxySPEX achieves comparable performance to SPEX with roughly 10x fewer ablations by leveraging hierarchical structure. The methods are evaluated on feature attribution (sentiment analysis), data attribution, and mechanistic interpretability tasks, outperforming marginal methods like LIME at long context lengths.
Related guides (3)
Related events (8)
LLM-augmented XAI framework with mutual feature interactions for network operations
A new arXiv paper proposes a framework combining LLMs with SHAP-based explainability, augmented by mutual feature interaction data, to generate natural language explanations for AI/ML models used in network operations. The approach is validated on an optical quality-of-transmission estimation task with human evaluators, showing 12.2% and 6.2% improvements in explanation usefulness and scope over a SHAP-only baseline, with 97.5% correctness. The work targets the gap between technical XAI outputs and actionable insights for non-specialist network operators.
SPECTRA: Synthetic IR Test Collections with Relevance Oracles and Controlled Distractor Diagnostics
SPECTRA is a reproducible framework for generating synthetic information retrieval test collections, separating latent topical structure, surface text realization, and query intent generation to produce deterministic relevance oracles without human annotation. A Python prototype generated corpora up to 60,000 documents at roughly 12K–14K documents per second, with graded relevance labels for 96 queries. Controlled distractor experiments showed BM25 nDCG@10 degrading from 1.00 at 2% distractors to 0.43 at 36%, demonstrating the framework's utility for exposing retrieval system failure modes before expensive real-world collection construction. The authors position SPECTRA as a diagnostic complement to Cranfield/TREC-style evaluation rather than a replacement for human judgment.
ModSleuth: Agentic system audits invisible dependency graphs in modern LLM training pipelines
Researchers introduce ModSleuth, an agentic system that recursively reconstructs LLM dependency graphs from public artifacts, recovering 1,060 source-verified dependencies across four major LLM releases. The system formalizes direct and indirect dependencies and operation-centered relationships to handle fragmented, inconsistent documentation. Applied at scale, the resulting graphs expose multi-hop license obligations, train-evaluation coupling, and discrepancies between released and training-time artifacts — issues that are practically invisible to manual auditing.
AgentSpec: A modular framework for controlled composition and analysis of embodied LLM agent scaffolds
AgentSpec is a new modular specification framework that represents embodied LLM agents as typed compositions of reusable policy components with standardized interfaces across perception, memory, reasoning, reflection, action, and learning modules. The framework enables controlled swapping and recombination of components, instantiated across four benchmarks (DeliveryBench, ALFRED, MiniGrid, RoboTHOR). Key findings include that agent performance is governed by scaffold compatibility and interaction effects rather than isolated module strength, and that RL-trained policies compose best when optimized with deployment-time scaffold structure. Code, baselines, and an interactive playground are publicly released.
AXPO: Agent Explorative Policy Optimization Addresses Thinking-Acting Gap in Multimodal Agentic Reasoning
This paper identifies a structural asymmetry in agentic reasoning called the 'Thinking-Acting Gap,' where tool use is attempted in only ~30% of rollouts under standard RL training (GRPO), and all-wrong tool-using subgroups suppress learning signals. The authors propose AXPO (Agent eXplorative Policy Optimization), which fixes the thinking prefix and resamples tool calls for all-wrong subgroups, combined with uncertainty-based prefix selection. Evaluated across nine multimodal benchmarks on Qwen3-VL-Thinking at multiple scales, SFT+AXPO outperforms SFT+GRPO by +1.8pp on both Pass@1 and Pass@4 at 8B, with the 8B model surpassing the 32B baseline on Pass@4 using 4× fewer parameters.
LEAF-X: Entropy-guided explainability framework for transformer-based ASR models
Researchers introduce LEAF-X (Listening with Entropy-guided Attention for Faithful explainability), a model-intrinsic XAI framework for transformer-based automatic speech recognition systems like Whisper. The method combines entropy-guided attention weighting, multi-layer attention rollout, and optional causal ablations to produce sparse token-to-frame attributions. Evaluations show 32% improved faithfulness and 35-39% stronger locality/sparsity compared to perturbation-based explainers and raw attention maps, enabling more auditable ASR.
SAEs as Stethoscopes: Interpretability-Guided Layer Selection for Task Vector Model Editing
This paper evaluates a Sparse Autoencoder (SAE)-guided model editing pipeline for mathematical reasoning on Gemma-3-4B-IT, finding that projecting task vectors onto SAE feature subspaces discards ~97% of modification energy due to geometric misalignment between activation-space SAE directions and weight-space task vectors. The authors reframe SAEs as diagnostic tools ('stethoscopes') rather than intervention filters ('scalpels'), using SAE-derived specificity scores to identify which layers to inject unfiltered task vectors into. This approach improves Number Theory accuracy from 29.6% to 39.4% on Minerva Math (p=0.0007), with 5 of 7 math subjects significantly improved and none degraded. The method is fully deterministic and adds no inference cost.
Self-Policy Distillation via Capability-Selective Subspace Projection
This paper introduces Self-Policy Distillation (SPD), a self-distillation method for LLMs that requires no external signals such as correctness filters or reward models. SPD extracts a low-rank capability subspace from the model's own gradients on correctness-defining tokens, then projects KV activations into this subspace during self-generation to isolate task-relevant signal from stylistic noise. Experiments across code generation, math reasoning, and QA show up to 13% improvement over prior signal-free self-distillation methods and 15% better out-of-domain generalization.