Natural Language Processing news and analysis on Web Pulse: 3984 curated articles tracking the latest Natural Language Processing developments, tools, and research, updated continuously from vetted sources.
A developer built an AI-powered assistant using OpenAI's GPT API to automatically generate Playwright test cases from natural language requirements. The system parses user stories or acceptance criteria and suggests Play…
Over 30% of new arXiv submissions now read as AI-written, according to a study that scored 12,750 papers using a detector calibrated to a 0.4% false-positive rate on pre-LLM text. The flagged share rose from 0.4% in 2021…
Clay engineer Alex Komoroske published an internal policy on acceptable AI writing use, arguing that as of 2026, AI models cannot produce lossless transformations of natural-language text and that unedited AI output risk…
A study of 16 volunteers found that conversations with AI-generated 'ghosts' of deceased loved ones were often comforting, even when the chatbot invented facts, according to information scientist Jack Manning of the Univ…
World Vision Zambia MERL manager Chanda Mwila won funding to develop a multi-modal AI-enabled app for Community Health Workers in rural Zambia that uses NLP and computer vision for diagnosis and triage across health, nut…
AI-powered search engines, including Google AI Overviews, Claude, ChatGPT, and Perplexity, now personalize answers for individual users based on context, location, and behavior, moving beyond traditional universal rankin…
PaddlePaddle requests adding the Real5-OmniDocBench framework and benchmark entry to Hugging Face's Hub Benchmark evaluation system. The benchmark evaluates document parsing robustness under five real-world acquisition s…
A nonprofit called Current AI, backed by France, DeepMind, Salesforce, and the Ford and MacArthur foundations with over $400 million in commitments, launched at France's AI Action Summit in early 2025 to build a free, op…
Apple is testing a hidden popover interface for Siri AI in the macOS 27 Golden Gate beta that surfaces Writing Tools and contextual actions when text is selected, as spotted by a Reddit user. The feature, disabled by def…
Perplexity offers three advanced techniques to improve AI search results: using search operators like site: and filetype: to narrow queries, creating Spaces with custom instructions for repetitive tasks, and prompting th…
A new study from arXiv finds that external toxicity tools are unreliable for code-mixed and multilingual abuse detection, proposing ToxGate, a trust-fusion head that conditions auxiliary signals on encoder representation…
Researchers propose AdaLook, an adaptive lookahead framework for masked diffusion language models that dynamically determines rollout depth based on candidate-score variance, achieving a better accuracy-decoding steps tr…
A new arXiv preprint (2607.15715v1) investigates whether agentic components like reflection and memory in LLM agents produce observable and controllable improvements over fixed workflows for information extraction. The s…
A study using the ASAP 2.0 dataset found that GPT-5 mini achieved the highest agreement with human ratings for automated essay scoring, while GPT-5 produced the strongest summarization quality, revealing trade-offs among…
Researchers introduced DrawingVQA, the first benchmark for evaluating multimodal large language models (MLLMs) on real-world construction drawings, featuring 33 'Issued for Construction' drawings and 92 expert-curated qu…
Researchers introduced EpiNarrate, an agentic framework for generating grounded public health narratives from epidemiological projections, which separates structured numerical reasoning from natural-language generation t…
A new study measuring how much formal semantic structure explains human label variation in natural language inference finds that it accounts for only 3.3 to 3.6 percent of entropy variance, with a group-level effect show…
A new opinion paper from arXiv argues that large language model (LLM) self-explanations are often plausible and actionable but may lack faithfulness to the model's true reasoning. The authors identify limitations in stan…
A new study from arXiv finds that test-time adaptation (TTA) can improve zero-shot 3D CT vision-language models only under specific conditions, and introduces CARVE (Cardinality-Aware Retained-View Entropy) as the first …
A new study from researchers evaluating large language models as unified multimodal learners for clinical prediction finds that converting all patient data into a single natural language sequence and fine-tuning a pretra…