Tag
The paper proposes ModelLog, a declarative probabilistic framework for evaluating language models by defining semantic constraints over token predictions, linking evaluation to learning through shared semantics.
Cortex is a web application that uses artificial intelligence and natural language processing to assist researchers in qualitative content analysis, providing support for pre-analysis and categorization tasks.
The paper introduces a weakly supervised framework using large language models to extract dataset mentions in forced displacement and FCV documents, achieving high accuracy with limited labeled data.
WrenAI is an open-source generative BI engine that allows users to ask questions in natural language to generate SQL, charts, and analysis results, and it features a semantic layer to enhance accuracy.
Persimmon is introduced as the first large-scale model designed to realistically simulate how people talk and interact.
This paper proposes Prototype-Mediated Process Supervision (PMPS) for latent chain-of-thought reasoning, achieving token compression and accuracy improvements over explicit CoT methods.
The paper proposes an end-to-end pipeline for natural language to PDDL translation using LLMs with iterative repair and comprehensive evaluation, highlighting gaps between operational success and semantic faithfulness in benchmarks.
This paper presents a controlled inversion test to examine whether recognizing fact-preserving news framing can lead to its reversal.
Introduces SWRouter, a similarity-contractive window router for multi-turn large language model conversations, addressing challenges in context segmentation and routing evaluation.
This paper proposes using semantic uncertainty derived from large language models to anticipate transition relevance places in spoken turn-taking, showing improved performance over baselines in dialogue systems.
This paper introduces a benchmark dataset for Bangla idioms and evaluates recent large language models on idiom-related tasks, revealing substantial variability in performance across models.
FrameBench is a new benchmark for evaluating whether large language models can distinguish context-dependent frame-semantic interpretations of verbs, constructed for English and Japanese using FrameNet resources and released with code.
MineTRACE is a web-based evidence-grounded interactive reasoning system that integrates geochemical, geophysical, and geological data to provide transparent mineral prospectivity scores and natural language interaction, supporting efficient and verifiable mineral exploration.
This paper explores how persona prompting with different attribute selection methods affects the alignment of large language models with human responses in social surveys. It finds that effectiveness depends on human response variation and the choice of attributes.
MultiGhostBench is a multilingual benchmark for long-form LLM-generated text attribution under distribution shifts, featuring 928 books in six languages and highlighting performance degrades and no single method consistently best across settings.
This study proposes a two-stage framework for sentence-level depression symptom recognition using candidate generation and definition-guided verification, achieving best accuracy and F1 scores among evaluated methods.
The article explores Pangram, an AI startup that detects AI-generated text, its role in publishing scandals, and questions about the trustworthiness of its detection model.
This article provides a detailed explanation of the Transformer architecture, covering attention mechanisms, QKV, and residual connections, while tracing its historical development from N-gram to LSTM.
This paper investigates the curse of multilinguality in lexical normalization, finding that training a single model on multiple languages leads to decreased per-language accuracy, with optimal performance when languages are trained in small groups.
CUDA-Harness is a framework that uses agentic techniques to generate and optimize CUDA kernels from natural language descriptions, addressing challenges in Text2CUDA by connecting high-level semantics with low-level implementation and verification.