LLMs
Latest breakthroughs in Large Language Models
Articles
OpenClaw and Ollama in Agentic AI: Toward Fully Autonomous and⭐8
OpenClaw and Ollama enable autonomous, scalable AI agents via a layered architecture, with validation showing performance gains from integration.
DoTime: A Synthetic Benchmark Generator for Interventional and⭐7
DoTime offers open, scalable synthetic data for evaluating causal time-series methods, with continuous-time interventions and counterfactual sampling.
Regularizing Modality Contribution Drift in Multimodal Continual⭐8
Multimodal continual learning models suffer from Modality Contribution Drift (MCD). We propose CMCDR regularization with replay-based and replay-free
Recursive Transformers for Semiconductor Thermo-Mechanical Reliability⭐10
Recursive transformers with weight sharing deliver accurate, efficient surrogate modeling for semiconductor thermo-mechanical reliability, balancing
AI-Assisted Pre-Review of Open-Source Software Submissions: An⭐9
BOSC 2026 tested AI-assisted pre-review for software submissions using agentic skills and automated builds, helping reviewers gather evidence while
Prompt Chaining in Practice: A Case Study in Automated Scholarly⭐9
Prompt chaining boosts reliability to 100% vs 50% for single-shot LLMs in scholarly report generation, with higher ROUGE-L F1 (0.507 vs 0.486).
Meta-Learned Reward Shaping for Reinforcement Learning from⭐8
Meta-learned reward shaping (MeRLa) boosts LLM alignment with denser signals, achieving 90.8% win rate on AlpacaEval 2.0 and surpassing PPO, DPO.
Sim2Win: A Team-Agnostic, Event-Based Pre-Match Outcome⭐8
Sim2Win offers team-agnostic pre-match football predictions using event-based tactical profiling and machine learning, enabling actionable insights without team...
Emergent Sparsity in Frozen Random CNN Feature Extractors for⭐7
## Emergent Sparsity in Frozen Random CNN Feature Extractors for Deep Reinforcement Learning **Authors:** Scott M.
DuplexGen: Adaptive Synthesis of Human-AI Turn-Taking Dialogues⭐8
DuplexGen uses human preference calibration to adaptively synthesize turn-taking dialogues, outperforming uncalibrated models across six cooperative and competi...