Natural Language Processing

Researchers introduced CORTEX, a new method for detecting hallucinations in AI-generated text at the token level within Retrieval-Augmented Generation (RAG) systems. By analyzing internal model representations with and w…

A new unified AI model integrates textual reasoning and graph processing into a single masked diffusion language model, outperforming graph neural networks, graph transformers, and LLM-based baselines by up to 3.9 points…

A new article explains that encoder and decoder models like BERT and GPT share the same underlying architecture, with the key difference being which tokens each model is allowed to attend to during processing.

Ragit, a new command-line tool, enables users to chat with any folder of documents using a local LLM via Ollama. It implements Retrieval-Augmented Generation (RAG) by indexing files, creating embeddings, and retrieving r…

A developer explains how prompt engineering can significantly improve the quality of AI-generated code and responses. The guide covers key techniques such as defining roles, providing context, and using step-by-step prom…

AI is transforming data engineering by acting as a super compiler that translates natural language into executable code, potentially making traditional languages like SQL obsolete. The target language for this translatio…

AI is revolutionizing simulation model discovery using transformer-based embeddings and retrieval strategies, enabling natural language queries to find relevant models. A recent study shows open-source embedding models a…

Researchers developed HS2C, a method that compresses graph data by leveraging homophily to improve large language models' reasoning accuracy. Tested on 10 node-level benchmarks, HS2C enhances compression and accuracy, ma…

Cohere and LG CNS have unveiled LuckyStar 111B, a hybrid reasoning model designed to enhance AI agility for Korean-English enterprise tasks. Built on Cohere's Command A model, it uses multilingual fine-tuning and reinfor…

BiRG-LoRA, a single-adapter rank-gated method, achieves 69.31% macro-average accuracy on medical QA benchmarks while using 28.1% fewer parameters than MoELoRA, offering a more efficient and accurate AI solution for healt…

Researchers introduced Outcome Reward Models (ORMs) for Text-to-SQL verification, using semantic scoring to outperform traditional heuristics. Their GradeSQL framework automates candidate generation and labeling, achievi…

Researchers testing GPT-2 style models on 'impossible' languages found that while the models showed gradual decline in grammatical sensitivity, they struggled significantly with generative production, failing to produce …

Researchers introduced TheraJudge and TheraAgent, a framework that uses multi-agent systems and human-aligned evaluation to improve therapeutic quality in mental health support LLMs, achieving a +0.43 improvement in clin…

Researchers released a benchmark for Arabic-Russian scientific translation, including a 27,000-sentence parallel corpus and fine-tuned multilingual models. The Qwen2.5-7B model achieved BLEU 23.15, outperforming zero-sho…

Researchers introduced Contrastive Reflection, an iterative prompt-optimization framework for agentic information retrieval workflows, which uses error-anchored behavioral slices and contrastive examples to propose targe…

Researchers introduced a Bangla event detection benchmark with 9,979 annotated sentences across clean text, ASR transcripts, and corrupted text. They found encoder models perform better on clean text but degrade under no…

Researchers introduced Indi-RomCoM, a benchmark for evaluating large language models on Romanized code-mixed Indic-English instructions. Testing across seven tasks and four languages showed that LLMs consistently underpe…

Researchers introduced a framework using generative AI agents to automate black-box audits of personalization algorithms, deploying 1,120 agents on X after the 2024 U.S. election. They found that X's algorithmic feed amp…

Researchers introduced GradeSQL, a framework for training Outcome Reward Models (ORMs) to improve test-time verification in Text-to-SQL tasks. ORM-based selection outperformed execution-based Best-of-N and Majority Votin…

Researchers propose a reference-based evaluation protocol for prosody and rhythm in speech-to-speech AI agents, using 4000+ hours of dyadic English conversation to create matched reference regimes for metrics like F0 and…