LLMs
Latest breakthroughs in Large Language Models
Articles
QFoldAgent: An Autonomous Quantum Optimization Multi-Agent⭐7
QFoldAgent uses a multi-agent system with quantum optimization to improve protein structure prediction, reducing RMSD and boosting structural validity in 5-resi...
SeT-Diff: Semantic Foundation Models for HPC Telemetry and Time-Series⭐9
SeT-Diff introduces the first foundation model for HPC telemetry, using semantic conditioning and diffusion to achieve 0.047 MAE reconstruction, zero-shot
CausalGate: Causal Importance Distillation for Transformer⭐8
CausalGate uses causal importance distillation to prune transformer modules, reducing compute and latency in LLMs without runtime overhead.
CORVUS: Context Optimization and Reduction Via Underlying⭐8
CORVUS optimizes LLM coding agents by decoupling file reads from history, reducing tokens by 9-50% and reasoning cycles up to 37% while maintaining accuracy.
Semalith v1.4: A Calibrated 184M Safety Classifier Achieving⭐7
Semalith v1.4 is an 184M safety classifier achieving top prompt-injection detection with 44x fewer parameters than Llama-Guard-3-8B, plus zero false positives.
Evaluating the Impact of Reviewer Guideline Design on LLM-Based⭐6
Study compares LLM-based peer review using official vs. reviewer-imitating guidelines, finding official guidelines align best with human judgments.
MioFFAn: An Annotation Software for Formula Formalization with⭐9
MioFFAn is an open-source annotation tool for formula formalization, enabling dataset creation with customized taxonomies and partial LLM automation.
On the Depth Scalability of Logic Gate Networks⭐9
This paper identifies why Logic Gate Networks fail to scale with depth and introduces Input-Anchored Logic Gate Networks, achieving consistent accuracy gains be...
Cloud-Native Evaluation-as-a-Service: A Microservices⭐8
A cloud-native architecture for scalable AI monitoring using six stateless microservices, with conformal guarantees and real-time drift detection.
A Consensus-Based Framework for Relative Preference Evaluation⭐7
A consensus-based framework for LLM evaluation measures relative preference among model-generated responses instead of absolute correctness, using a voting pane...