Natural Language Processing news and analysis on Web Pulse: 3923 curated articles tracking the latest Natural Language Processing developments, tools, and research, updated continuously from vetted sources.
A new arXiv preprint (2608.19299v1) reports that large language models can generate operationally realistic air traffic control transmissions, but only with lightly constrained prompts and in-context examples. In experim…
A study submitted to arXiv on 2 Jun 2026 compares transformer models BERT, RoBERTa, and BART for text summarization, examining their architectures, pretraining strategies, and suitability for extractive and abstractive t…
A virtual assistant called ATHENA, developed for the Society of Petroleum Engineers, improved productivity and performance equality among 75 professionals on well-planning tasks compared to a state-of-the-art RAG baselin…
Researchers present SNAIL, a hybrid named entity recognition framework that automatically identifies bioinformatics software and database names in biomedical texts, outperforming existing methods including bioNerDS2, Cha…
A new study from arXiv (2608.19211v1) finds that four audio-language models preserve prosodic information in their audio paths and late LLM states but fail to express it in final responses, with targeted hidden-state int…
A new arXiv study (2608.19558v1) finds that confidence rankings for financial named entity recognition (NER) change under domain shift, with whole-output probability being the strongest in-domain error detector but deter…
Researchers from an unnamed institution report that a contrastive decoder can extract open-vocabulary word-level information from non-invasive EEG during silent reading, achieving reliable above-chance word-grouped top-1…
Cosine similarity, the default metric in most RAG pipelines and vector databases, discards vector magnitude, which can carry critical semantic information such as confidence, specificity, or negation, leading to failures…
A developer used AI-assisted coding to build a client-side Keyword Density Analyzer that runs entirely in the browser, addressing privacy concerns with third-party SEO tools. The tool leverages regex sanitization and pur…
Shopify Engineering implemented gisting to compress the Sidekick GraphQL agent's system prompt from ~6,000 tokens to ~1,500 gist tokens (a 4:1 reduction) without losing prediction quality, cutting median time to first to…
Hume AI's new research introduces three tests to quantify benchmark optimization in speech recognition, finding that several of 11 open-source ASR models reproduced benchmark transcripts from VoxPopuli and LibriSpeech ev…
Pipe Language released pipe-docs, a semantic search and RAG MCP server that turns documentation and Go/Pipe source code into a searchable index for AI IDEs, offering seven tools with zero dependencies and published on th…
Between 16th and 20th August, OpenAI replaced ChatGPT's search tool call from JSON to a compact query language with freshness windows, domain targeting, and verticals for products, places, images, and widgets, as observe…
OpenAI's ChatGPT search now uses the site: operator at scale, allowing users to restrict search results to specific domains, as announced by Simon Willison on his Weblog on August 20, 2026. This enhancement improves the …
A technical blog post outlines a pipeline for analyzing political communications using AI, including PII stripping with Presidio, embedding with mixedbread-ai/mxbai-embed-large-v1, fine-tuning a DeBERTa-v3-large classifi…
Superwhisper released S1-mini, a 462 MB open-weights text normalizer fine-tuned from Qwen/Qwen3-0.6B, which turns raw ASR transcripts into clean written text with 94.8% token accuracy on a held-out set of 7,519 cases. Th…
Vale, an open-source command-line linter for prose, enforces consistent editorial style across documents in 12 formats entirely offline, with no runtime and parallel file linting. Written in Go and MIT-licensed, Vale is …
Google Home is rolling out an August 20 update that improves Spotify command recognition, changes broadcasts to a two-step process, and adds voice control for setting smart light color temperature to specific kelvin valu…
Anthropic has introduced a new output style for its Claude AI model, called 'Plain English,' which instructs the model to lead with the answer, use common vocabulary, and keep responses concise. The style guide, publishe…
Open-source and commercial 'unslop' tools that rewrite AI-generated text to sound human are emerging as a new software category, with at least three MIT-licensed projects on GitHub, a pip install unslop on PyPI, and mark…