Natural Language Processing news and analysis on Web Pulse: 5427 curated articles tracking the latest Natural Language Processing developments, tools, and research, updated continuously from vetted sources.
A developer built Greek Invented Words, a Kaggle benchmarking submission that sends 100 short Greek prompts to 15 language models and scores answers without an LLM judge, using native-speaker verdicts from Apollon Labs t…
Scrieb launched as an offline meeting transcription, diarization and summarization tool that runs AI models locally on a laptop, keeping recordings on-device with no cloud upload or internet connection required. The tool…
A developer built a 40-line Python detector that scores text as AI-generated using burstiness (coefficient of variation in sentence length) and repeated n-grams, and reported that it fails in three ways: short texts unde…
Developer ahmedhmam1994 released VoxScribe, a free open-source Windows voice dictation app that transcribes speech locally using faster-whisper on the CPU and types the result directly into the focused window. The app is…
ElevenLabs released Eleven v4 and Eleven v4 Turbo, a new speech-generation architecture with a low-latency variant that has a median inference latency of roughly 100 ms. Eleven v4 Turbo is priced at $11 and Eleven v4 at …
A developer embedded all 6,142 Y Combinator companies with live domains across 48 batches (Summer 2005 through Fall 2026) using text-embedding-3-large at 1024 dimensions and found that 2025 batches are roughly 2.5x more …
A developer's technical breakdown of AI voice agents explains that the systems are built from three streaming models — speech-to-text, a language model, and text-to-speech — wrapped in telephony, and that end-to-end late…
A developer based in Vietnam built an AI sourcing agent that converts free-text product briefs into business Vietnamese factory requests using three separate model calls, with the first two steps running as live model ca…
A new arXiv paper (2609.35869v1) measures the compression cost of pre-tokenisation boundary rules, finding that on English Wikipedia such boundaries increase the optimal token count by 28.3–36.8%. The authors report that…
Researchers introduced Structured Matrix Attention (SMat-Attention), a long-context sequence modeling method that uses a family of causal masks with structured long-range routing whose row supports have VC-dimension d, a…
A new post-training method called GapFT improves single-sample decoding accuracy of Llama-3.1-8B by 14.4 points on LogiQA 2.0 and 13.9 points on ReClor over the source model, according to an arXiv paper (2609.35793v1). G…
A new arXiv paper (2609.35804v1) reports that prompt perturbations can mitigate bias and hallucination in some large language models, contrary to previous studies. The study found Claude 3 was more effective across the t…
TRACE, a tree-relational enhancement framework for oncology large language models, improved label-free evaluation and supervised fine-tuning across ten oncology classification tasks and the MedQuAD CancerGov QA benchmark…
Researchers proposed TagPR, a framework that adds semantic tags to a large language model's reasoning process for step-by-step guidance on personalization tasks, according to an arXiv paper (2509.23140v2). TagPR generate…
A neurosymbolic router that learns a deterministic finite automaton via the L* grammatical inference algorithm achieved 100% routing accuracy and 98.3% overall accuracy on 100 untested prompts from DeepMind Mathematics, …
Researchers released GeoOutageBench, a benchmark for evaluating LLM-based geospatiotemporal knowledge-graph question answering (KGQA) on multimodal power outage and resilience analysis, with code, data, and results avail…
A new arXiv paper (arXiv:2609.35856v1) proposes using generative large language models with weak supervision to classify Economic Policy Uncertainty (EPU) in news articles, replacing keyword-based methods that produce hi…
A team of researchers has developed an Optical Character Recognition (OCR) model that combines a deep learning model with image preprocessing techniques to extract information from Korean-language invoices, achieving an …
Researchers introduced FD-VAD, an ASR-free streaming endpointer that maps bounded causal audio windows directly to Continue/Stop decisions for full-duplex voice interaction, combining a frozen speech encoder with a light…
A lightweight module called Sieve, paired with a costly LLM dubbed Sage, improved Retrieval-Augmented Language Model accuracy by up to 69.4 percentage points and Macro-F1 by 55.2 percentage points over one-stage baseline…