AINeutralarXiv – CS AI · Jun 36/10
🧠Researchers introduce a geometric decomposition framework to understand how prompting reshapes internal representations in large language models and vision-language models without weight updates. Testing across multiple models and datasets reveals that prompts consistently reorganize representations toward task structures, with cross-dimensional linear mixing (affine transformations) emerging as a key mechanism for prompt-driven behavior.
AINeutralarXiv – CS AI · Jun 26/10
🧠Researchers introduce Partial Information Decomposition (PID), a framework for analyzing how multimodal language models integrate vision and language inputs by separating unique, redundant, and synergistic contributions. The analysis reveals distinct modality-use patterns across task types and identifies visual dominance as a bottleneck in audio-visual fusion systems.
AINeutralarXiv – CS AI · May 296/10
🧠Researchers discovered that large language model failures in clinical triage stem from output formatting constraints rather than deficient medical knowledge. Using sparse autoencoders to analyze model internals, they found medical features activate identically across free-text and multiple-choice formats, but scaffold features drive incorrect decisions at the decision token, suggesting the models possess clinical understanding but struggle with constrained response structures.
AINeutralarXiv – CS AI · May 286/10
🧠Researchers have developed methods to identify which attention heads in Large Language Models are responsible for specific reasoning steps, revealing that only ~3% of heads handle factual retrieval while higher layers coordinate multi-step reasoning algorithms. This work provides insights into how LLMs learn logical reasoning from limited demonstrations and could improve model interpretability and design.
AINeutralarXiv – CS AI · May 286/10
🧠Researchers discovered that large language models develop geometric structures in their internal representations that mirror human perceptual organization across domains like color, pitch, and emotion, despite training only on text. These perceptual geometries emerge transiently in intermediate layers, providing new insight into how LLMs develop human-like conceptual understanding without direct sensory supervision.
AINeutralarXiv – CS AI · May 286/10
🧠Researchers identify specific attention heads in large language models responsible for cultural binding—associating cultural items with appropriate identities. Through mechanistic interpretability analysis, they find that steering these heads can improve cultural differentiation accuracy by 1-3 percentage points, revealing that models possess far more cultural knowledge than they actively use.
AINeutralarXiv – CS AI · May 286/10
🧠Researchers present the Integrated cross-Architecture Reasoning (IAR) framework, a novel methodology for interpreting how large language models perform reasoning tasks by combining multiple analytical probes—bandwidth-calibrated Mutual Information Peak, Deep-Thinking Ratio analysis, and Jaccard stability metrics—across model layers and architectures. Testing on Qwen and Llama models across mathematics, code, logic, and common sense domains demonstrates that this multi-metric approach provides more reliable insights into LLM reasoning patterns than single-probe methods.
🧠 Llama
AIBullisharXiv – CS AI · May 286/10
🧠Researchers propose TELLME, a novel method to improve transparency and monitorability of large language models by enhancing their internal representations rather than relying solely on external monitoring tools. The technique demonstrates consistent improvements in detoxification tasks across multimodal datasets and model architectures, addressing the fundamental challenge that chain-of-thought explanations fail to accurately reflect LLMs' actual decision-making processes.
AINeutralarXiv – CS AI · May 286/10
🧠Researchers studying DeepSeek-V3 discovered that Large Language Models encode syntactic and semantic information in mathematically separable, linear patterns within their hidden layers. By averaging representations of sentences with shared structure or meaning, they created 'centroids' that capture significant linguistic information, revealing that syntax and semantics are processed through distinct, partially decoupled mechanisms across different layers.
AINeutralarXiv – CS AI · May 126/10
🧠Researchers present causal evidence that large language models learn in-context through dual mechanisms combining genuine structure inference with local pattern-matching, rather than relying on either approach alone. Using graph random-walk tasks and activation patching techniques, they demonstrate that LLMs simultaneously encode multiple competing graph topologies in orthogonal representational subspaces and show that late-layer circuits causally drive graph-preference predictions.
AINeutralarXiv – CS AI · May 116/10
🧠Researchers introduce Hidden-state Driven Margin Intervention (HDMI), a new probe-free technique for causal probing in large language models that directly manipulates hidden states without training auxiliary classifiers. The method achieves higher reliability than existing approaches by balancing completeness and selectivity across multiple benchmarks.
🧠 Llama
AIBearisharXiv – CS AI · May 96/10
🧠Researchers discovered that failure modes in medical LLMs (specifically 'Overthinking' behaviors) are linearly decodable in hidden states yet cannot be corrected through fixed linear steering interventions, revealing fundamental representational entanglement that limits straightforward correction approaches. However, the decodable failure signals enable effective selective abstention for reliability estimation.
AINeutralarXiv – CS AI · May 96/10
🧠Researchers introduce HyperLens, a high-resolution analysis tool that measures cognitive effort in large language models by tracking confidence trajectories across transformer layers. The study reveals that complex tasks consistently require higher cognitive effort and identifies how standard fine-tuning can paradoxically reduce model performance by decreasing necessary cognitive investment.
AINeutralarXiv – CS AI · May 96/10
🧠Researchers propose a novel black-box confidence estimation method for chain-of-thought reasoning that measures trajectory convergence rather than relying on expensive sampling. Testing across multiple benchmarks and AI models shows significant improvements over self-consistency baselines while requiring only 4 samples instead of 8, with potential applications for safer API-based AI deployment.
🧠 GPT-5🧠 Claude🧠 Sonnet
AINeutralarXiv – CS AI · May 96/10
🧠Researchers propose Adaptive Elastic Net Sparse Autoencoders (AEN-SAEs) to solve feature starvation in neural network interpretability tools. The method combines L2 and adaptive L1 regularization to create a mathematically stable sparse coding system that improves feature extraction in large language models without requiring complex workarounds.
🧠 Llama
AINeutralarXiv – CS AI · Apr 206/10
🧠Researchers compare three explainability techniques—Integrated Gradients, Attention Rollout, and SHAP—for interpreting LLM decisions on sentiment classification tasks. The study reveals that gradient-based methods offer stability and interpretability, while attention-based approaches are faster but less predictive, highlighting critical trade-offs in choosing explanation methods for transformer models.
AINeutralarXiv – CS AI · Apr 206/10
🧠Researchers conducted a comparative study of how large language models trained with different fine-tuning methods (full fine-tuning, LoRA, and quantized LoRA) interpret code compliance tasks. The study reveals that full fine-tuning produces more focused attribution patterns than parameter-efficient methods, and larger models develop distinct interpretive strategies despite performance gains plateauing above 7B parameters.
AINeutralarXiv – CS AI · Apr 206/10
🧠Researchers introduce AtManRL, a method that combines differentiable attention manipulation with reinforcement learning to improve the faithfulness of chain-of-thought reasoning in large language models. By training attention masks to identify which tokens genuinely influence model predictions, the approach demonstrates that LLM reasoning traces can be made more interpretable and transparent.
🧠 Llama
AINeutralarXiv – CS AI · Apr 206/10
🧠Researchers propose TPA (Token Probability Attribution), a new method for detecting hallucinations in Retrieval-Augmented Generation systems by attributing token generation to seven distinct sources rather than the traditional binary approach. The technique uses Part-of-Speech tagging to identify anomalies in how different linguistic categories are generated, achieving state-of-the-art detection performance.
AINeutralarXiv – CS AI · Apr 156/10
🧠Researchers demonstrate that large language models develop attractor-like geometric patterns in their activation space when processing identity documents describing persistent agents. Experiments on Llama 3.1 and Gemma 2 show paraphrased identity descriptions cluster significantly tighter than structural controls, suggesting LLMs encode semantic agent identity as stable attractors independent of linguistic variation.
🧠 Llama
AINeutralarXiv – CS AI · Apr 156/10
🧠Researchers introduce Safe-SAIL, a framework that uses sparse autoencoders to interpret safety features in large language models across four domains (pornography, politics, violence, terror). The work reduces interpretation costs by 55% and identifies 1,758 safety-related features with human-readable explanations, advancing mechanistic understanding of AI safety.
AINeutralarXiv – CS AI · Apr 156/10
🧠Researchers propose a novel framework treating Large Language Models as attention-informed Neural Topic Models, enabling interpretable topic extraction from documents. The approach combines white-box interpretability analysis with black-box long-context LLM capabilities, demonstrating competitive performance on topic modeling tasks while maintaining semantic clarity.
AINeutralarXiv – CS AI · Apr 156/10
🧠Researchers propose a method for large language models to handle ambiguous user requests by generating structured responses that enumerate multiple valid interpretations with corresponding answers, trained via reinforcement learning with dual reward objectives for coverage and precision.
AINeutralarXiv – CS AI · Apr 146/10
🧠Researchers investigate how large language models represent emotions in their latent spaces, discovering that LLMs develop coherent emotional representations aligned with established psychological models of valence and arousal. The findings support the linear representation hypothesis used in AI transparency methods and demonstrate practical applications for uncertainty quantification in emotion processing tasks.
AIBullisharXiv – CS AI · Mar 126/10
🧠Researchers developed Causal Concept Graphs (CCG), a new method for understanding how concepts interact during multi-step reasoning in language models by creating directed graphs of causal dependencies between interpretable features. Testing on GPT-2 Medium across reasoning tasks showed CCG significantly outperformed existing methods with a Causal Fidelity Score of 5.654, demonstrating more effective intervention targeting than random approaches.
