Natural Language Processing news and analysis on Web Pulse: 3972 curated articles tracking the latest Natural Language Processing developments, tools, and research, updated continuously from vetted sources.
AI chatbots like ChatGPT, Gemini, and Claude often produce confidently wrong answers due to their underlying architecture of next-token prediction and reinforcement learning from human feedback (RLHF), which prioritize f…
Researchers at the Okinawa Institute of Science and Technology (OIST) created a virtual robot with a brain-inspired neural network that learns language faster when driven by curiosity, achieving genuine language understa…
A new study from arXiv finds that large language models (LLMs) frequently generate ungrounded content when producing podcast transcripts from documents, with even GPT-4o showing significant unfaithfulness. The researcher…
A research team introduced ARI, a framework for historical document restoration that uses retrieval-augmented generation with large language models to restore illegible named entities requiring external historical knowle…
A new study analyzing self-harm representations in large language models finds that self-harm information crystallizes in the final 3-7% of network layers across four models, with Gemma-3-4B representing the contrastive …
Researchers propose TextSLIP, a medical vision-language pretraining framework that augments CLIP with intra-modal text contrastive learning to improve fine-grained semantic supervision for radiology report generation. Pr…
Researchers propose EVL-MCoT, an enhanced vision-language multi-chain-of-thought approach for harmful meme detection, achieving promising results on the HatefulMemes and MultiOff datasets. The method addresses limitation…
A new pipeline using large language models to extract grammatical rules and lexicons from grammar books generates synthetic parallel corpora for fine-tuning machine translation models, improving over seed-data baselines …
Researchers propose Visual Saliency Steering Distillation (VSSD) to improve multimodal chain-of-thought reasoning in small models by using attention maps from multimodal large language models to generate perturbed images…
Researchers from the University of Moratuwa have released grapheme-kit, an open-source Python library that computes lexical distance, similarity, and evaluation metrics on grapheme clusters rather than Unicode code point…
HSBC is hiring more than 100 artificial intelligence specialists for its first global AI centre of excellence in Singapore, set to launch later this year, and will also add 100 wealth relationship manager roles over the …
HyperNexus introduces Progressive Routing, a multi-layered system that reduces AI agent tool context overhead by 95%, from 50,000 tokens to 1,500 tokens, by semantically matching only relevant tools to the current task.
A step-by-step playbook for automating support tickets with AI, based on a documented case where a small team reduced weekly issues from 52 to 19 by eliminating root causes rather than speeding responses, recommends mapp…
New research from Amazon.com finds that leading language models, including versions of ChatGPT and Google Gemini, are prone to 'interpretive' transcription, often rewriting imperfect text into more plausible forms, with …
A curated index of over 30 free AI/ML textbooks has been published on GitHub by MarcosSete, with a weekly GitHub Action to validate all links and ensure they point to official author or publisher sites. The collection co…
Lexicon, a free, open-source desktop writing assistant that runs entirely locally, has been released on GitHub under an MIT license. The app provides grammar checking via LanguageTool and AI writing tools such as rewriti…
Finance One, an Australian non-bank lender, has deployed Icana.AI's CallCoach to analyze every customer conversation across its contact centers, replacing sample-based quality checks for about 2,000 calls per day. The fi…
Black Forest Labs released FLUX 3, a multimodal foundation model that jointly learns from images, video, audio and robot action prediction within a single architecture, generating clips up to 20 seconds with native audio…
US Neural built Mycelium, a decoupled semantic tool registry that replaces LangChain's AgentExecutor for sub-10ms local tool routing. The system achieves 9.6ms cold discovery latency and 70.7% family-level intent matchin…
A developer built SayChef, a voice-first recipe manager that converts natural speech into structured recipes. The biggest challenge was handling imperfect dictation—people pause, jump back, or change measurements mid-sen…