The Archive

PhyAI: unified inference engine for physical AI across evaluation, cloud RL, edge serving, and onboard deployment with single runtime.

VetScore evaluates veterinary QA outputs by assessing citation faithfulness and weighting claims by medical harm potential.

DiagLoop generates counterfactual training data from clinical guidelines to improve diagnostic LLM reasoning in low-data settings.

CausalOPD distills step-dependent causal reasoning into smaller models using online process distillation with first-wrong-step supervision.

Higher-order safety shields enforce multi-constraint safety (speed, force, jerk limits) for cyber-physical systems beyond binary state safety.

RAPO framework addresses catastrophic forgetting in continual multimodal LLM post-training via explicit dual-channel risk governance.

Study compares GPT-5.4, Gemini 3.1 Pro, and Claude Opus 4.6 peer reviews on 300 ICLR submissions against human reviewer alignment.

Modular generation-selection framework for factually consistent abstractive summarization under sentence budget constraints.

AutoSND uses tree search to discover heuristic policies for network dismantling by converting LLM execution feedback into structural guidance.

CILER models latent environment effects on user preferences for out-of-distribution recommendation using conditional identifiability.

Study evaluates zero-shot coordination robustness across independent algorithm implementations to assess practical agent alignment.

MuEvo uses LLMs to evolve multi-component heuristic ensembles for combinatorial optimization, addressing inter-component dependencies.

Study identifies spurious language priors in LLM token-level supervision during on-policy distillation and proposes mitigation.

POEM applies SO(2) feature rotation to handle periodicity drift in deep time series forecasting.

Closed-form analysis of cross-layer interactions under weight-space ablation in transformer models, validating on pretrained models.

ConformalShift demonstrates event-reordering attacks on adaptive ECG monitoring via conformal prediction without waveform modification.

Systematic study reveals gender bias in LLM-based fake news detection across six state-of-the-art models using LIAR benchmark.

Proposes security-oriented lifecycle model for LLM systems addressing provenance, signing, permissions, and decommissioning in critical infrastructure.

LoopMTP combines looped transformers with multi-token prediction to improve parameter-efficient reasoning via intermediate supervision.

Theoretical analysis of when activation patching and weight-space ablation agree in idealized models, with synthetic validation.

Machine-readable catalogue and OCR analysis of Tsiolkovsky personal archive (2,019 files, 51,008 scans) with handwriting classification.

Study proposes explicit modality reliability modeling for incomplete multimodal sentiment analysis across text, audio, vision.

Multi-teacher distillation method decouples language-specific knowledge in multilingual LLM-based automatic speech recognition.

Formal verification framework for LLM-agent systems operating on persistent relational data with business logic constraints.

On Tuesday, AI infrastructure company Runware announced the launch of its own modular data center called Sonic Inference Pod.

Google announces July 2026 AI updates; article lacks specific details on models, features, or benchmarks.

Offline reinforcement learning framework trained on 31.7k records to predict oncology clinical trial portfolios under uncertainty.

FraQ efficiently recompresses federated LoRA gradients in coordinate space, resolving aggregation mismatch in distributed LLM fine-tuning.

Survey of LLM applications across PDE workflows: discovery, numerical solver generation, diagnostics, and scientific feedback loops.

Byte-level language models enable exact knowledge transfer via shared output space and decompose language modeling from tokenization boundary prediction.