Natural Language Processing news and analysis on Web Pulse: 5612 curated articles tracking the latest Natural Language Processing developments, tools, and research, updated continuously from vetted sources.
Researchers at the University of Washington and Allen Institute for AI conducted the first fine-grained study of narrative features in Dolma, a 3-trillion-token open LLM pretraining corpus. They developed NarraBERT, a Ro…
A systematic experimental analysis of eight state-of-the-art Diffusion Language Models (DLMs) across eight benchmarks reveals that generation-time design choices strongly influence performance and computational efficienc…
Researchers at the Allen Institute for AI used training-data attribution to map which regions of the pretraining corpus support social-reasoning versus STEM-reasoning in the OLMo3-7B language model. They found that socia…
Researchers propose QueryGaussian, a training-free framework for open-vocabulary 3D instance retrieval that decouples semantic understanding from geometric representation, enabling efficient retrieval in city-scale scene…
FocusOnFun team's Vortex system achieved 90.5% in the Preliminary Round and 'Excellent' overall performance with 'Outstanding' QA results at the Ho Chi Minh City AI Challenge 2025. The multimodal video retrieval system i…
A developer describes building a complex 'loop engineering' pipeline in December 2024 to translate large Korean documents to English using local LLMs, but ultimately abandoned the project after weeks because the critic l…
A developer building agentic RAG systems found that the transition from simple RAG to agentic architectures introduces significant infrastructure challenges, including tool routing, infinite loops, and latency. The devel…
A startup has built the first AI runtime that enables models to shift behavior by reading the moment, addressing issues like drift, tone mismatch, and slop outputs by providing runtime insights into social norms and conv…
OpenAI is preparing a real-time voice mode for its Codex coding assistant, featuring a wake word, persistent session threads, and an animated pet avatar. The update borrows elements from ChatGPT's voice interface, signal…
Florida-based public safety technology company Aurelian has deployed an AI call-taker named Ava to answer police non-emergency phone lines, take reports, and route complaints. Aurelian CEO Max Keenan said roughly 70% of …
Apple's native iPhone apps like Notes and Voice Memos capture audio but require manual post-processing to create structured meeting notes, while AI-powered apps like Granola, Otter, and Minutes offer automated transcript…
Action items are specific tasks with a single owner, defined deliverable, and concrete deadline, structured as 'who does what by when.' AI notepads can automatically extract action items from meeting transcripts, allowin…
Zoom's free plan lacks native transcription, but five genuinely free third-party methods exist in 2026, including Granola which captures device audio without a visible bot. Most alternatives are free trials or require pa…
A new workflow using AI notepad Granola allows users to jot key decisions during meetings and have AI expand them into structured minutes using full transcript context, avoiding generic summaries. The method distinguishe…
Granola, an AI notepad, offers a method to record and summarize meetings across Zoom, Teams, Google Meet, and in-person settings without visible bots or stored audio. The tool captures device audio directly, deletes the …
Researchers conducting epistemic stress tests on closed large language models (LLMs) found that model breakdowns are not errors but ontological boundaries of predictive-text systems. The study observed distinct stability…
Florida tech firm Aurelian has developed an AI-powered call taker named Ava to handle routine non-emergency police calls in Los Angeles, where 911 response times are lagging. The system addresses chronic dispatcher short…
LectuLibre developed a chunking strategy to translate entire books using large language models while preserving narrative coherence. The pipeline parses documents into logical units like chapters, splits them at sentence…
A developer evaluated six LLM observability tools—Langfuse, Helicone, Arize Phoenix, LangSmith, Braintrust, and Laminar—for their ability to monitor the audio layer in voice agents, not just LLM calls. The tools are blin…
Mass General Brigham researchers developed BRIDGE, a multilingual benchmark that evaluates large language models on real-world clinical tasks, revealing significant gaps between AI performance on medical licensing exams …