LLMs
Latest breakthroughs in Large Language Models
Articles
A Large-Scale Dataset for Multi-Class Financial Newsโญ8
A 160,000-article Brazilian Portuguese financial news dataset with manual and LLM annotations, benchmarking multi-class classification via semantic similarity.
PRISM: PRIvileged Sibling-Guided Multi-Agent Topology Designโญ7
PRISM uses privileged sibling-guided multi-agent topology design to improve LLM collaboration. It achieves gains across six benchmarks over all baselines.
Value-Informed Schedule Trimming for Accelerated Sequentialโญ9
VISTA selects sparse SMC guidance timesteps in discrete diffusion via value-informed dynamic programming, cutting latency while matching full-guidance alignment...
Evaluating Model Capabilities After Benchmark Saturationโญ8
Saturated benchmark transcripts still reveal model capabilities, with pairwise judging predicting performance 5.3x better, though results can fail under distrib...
DeepUnfolding_Agent: A Classical_Quantum Adaptive Control APIโญ9
DeepUnfolding_Agent is a schema-driven FastAPI platform that decouples deep-unfolded adaptive control from training scripts, enabling reusable, safe, reproducib...
The World Won't Wait: Evaluating and Teaching LLM Agents Under aโญ8
RealtimeBench evaluates LLM agents under a live clock, showing deliberation latency hurts performance. TORSI, a two-timescale method, raises scores from 13.51 t...
Computer Science > Computation and Languageโญ9
Abstract page for arXiv paper 2610.18881: Following the Flow the Law Requires: Training Voice Agents to Transfer to Unseen Regulated Procedures in Debt Collecti...
Artificiality's First Exam: An LLM-Referenced C-Test forโญ9
This paper introduces the first Levin's Kt-based C-test using LLMs as representational machines to evaluate AI systems that may surpass human and current machin...
AutoIndexer: Flexible-Order Decoding by Training Causal Modelsโญ8
AutoIndexer trains causal models on chains of edits, enabling flexible-order decoding with token insertions, substitutions, and deletions for post-hoc context e...
Faithfulness Without Anchors: A Position Paper on the Structuralโญ9
Explainable AI faithfulness metrics have five major shortcomings that misalign feature importance explanations with model behavior, limiting trustworthiness qua...