The Archive

Layer-wise MBTI personality analysis of quantized LLMs reveals trait changes across precision levels.

ASR models reveal phonological traces of first language persist in pre-phonemic layers during second-language training.

SciMIF benchmark evaluates MLLMs' complex instruction-following across 22 tasks in 5 scientific domains.

PANDA framework transfers auxiliary modality signal to primary-modality medical models without paired data at inference.

Theoretical analysis of robust MDPs with linearly-defined uncertainty sets and ω-regular parity objectives.

LivingRAG adds writable experience store to Graph RAG, enabling reuse of verified reasoning across queries.

Lovable extends AI web app builder with MCP-powered capabilities for agent integration in SaaS workflows.

Spectral analysis detects AI-generated text by capturing signal fluctuations absent in confidence-based metrics, offering training-free scalability.

Study shows magnitude pruning degrades sparse autoencoders' interpretability of LLMs; activation-aware methods like Wanda preserve SAE robustness.

Audit of 51 physical AI models on 12 benchmarks reveals redundancy in benchmark suites, affecting model ranking reliability.

Multi-agent systems fail despite correct candidate answers; LLM judges improve selection under high candidate diversity.

On-policy self-distillation trains models on own generations with privileged teacher information; critical review of feasibility and scaling.

TAU-Agent combines retrieval and visual perception tools for traffic anomaly detection and explanation in video.

Cross-benchmark evaluation of nine fact-checking systems reveals poor generalization; no prior work spans full retrieve-verify pipeline across datasets.

Text-to-image models systematically fail on compositional prompts; human study of 651 images identifies defect patterns.

Particle’s new podcast intelligence platform transcribes and analyzes more than 130,000 podcasts, making their conversations searchable on the web and accessible to AI agents through an API and MCP.

Neural-operator surrogate replaces expensive 3D electromagnetic forward solver in Bayesian inversion for geophysical surveys.

Control variables mitigate omitted variable bias in deep networks; shows demographic covariates encode into predictions via shortcut learning.

Code World Model combines LLMs and video models to separate world dynamics from visual rendering for coherent, open-ended environment simulation.

Runtime verification framework monitors air traffic control procedures via formal methods to detect controller-pilot exchange failures and safety hazards.

Stage-isolation attack protocol traces GNN-based KGQA pipeline failures from entity linking through answer generation using adversarial perturbations.

SymTrace framework distinguishes causal repair from stochastic repair in LLM multi-agent systems, evaluating whether debugging methods fix root failures or exploit sampling variance.

Trajectory-based framework measures game-theoretic interaction structures (simultaneous, sequential, asymmetric) in real-world vehicle dynamics.

SAMpLE is an open-source SystemC-AMS framework integrating ML models as reusable Timed Dataflow components for virtual prototype simulation.

Quantum-inspired model captures heterogeneous, context-dependent driver behavior with continuous probabilistic dynamics and interpretability.

Controlled pretraining study compares romanization, IPA, and native orthography for multilingual LLM knowledge transfer across languages with different writing systems.

MetaSieve optimizes relational deep learning by using SQL to select metapaths and reduce GNN subgraph sampling costs on multi-table databases.

SCROLL framework forecasts multiple observables in stochastic dynamics via score-trained uncertainty with learned likelihood scaling per task.

Anima Anandkumar argues foundation models for physics simulation lag language models; applies ML to weather and fusion forecasting.

Proposes unified information bottleneck framework bridging attribution and counterfactual explanations for time series model interpretability.