Natural Language Processing

Researchers at ACL 2026 introduced a method to reuse LLM hidden states for classification tasks, training lightweight probes on token- and layer-selective representations to eliminate separate guard models. Their two-sta…

Researchers at the Association for Computational Linguistics found that large language models' numerical accuracy drops significantly when numbers are presented in underrepresented scripts or formats, even when the under…

Researchers introduced CoLabScience, a proactive AI assistant that uses a reinforcement learning framework called PULI to autonomously intervene in biomedical research discussions. The system, trained on a new benchmark …

Researchers at ACL 2026 found that entropy collapse in LLM reasoning, which undermines test-time scaling, is driven by premature overconfidence at a small set of critical tokens. They proposed SCOPE, a method that applie…

Researchers at Northwestern University and the University of Chicago developed a computational method to measure open codes in qualitative analysis, using LLM-enriched hierarchical clustering and four novel metrics (Cove…

Researchers at the 64th Annual Meeting of the Association for Computational Linguistics introduced a dual-task paradigm combining arithmetic and sentence comprehension to study how language models allocate cognitive reso…

Cerenovus, a Y Combinator Summer 2026 company founded by Harvard students, is building an AI-powered 'company brain' that aggregates and reasons over fragmented enterprise data from emails, Slack, documents, and spreadsh…

Large language models are stateless by default, but agents require memory to retain context across steps. A new technical guide identifies seven types of agent memory—working, semantic, episodic, procedural, retrieval, p…

A developer fine-tuned a tiny 0.6B-parameter Qwen 3 model to categorize household questions into metadata categories like pool, car, and HVAC. The baseline model achieved only 10% accuracy via prompting, but fine-tuning …

A new analysis argues that calling large language models like ChatGPT 'autocomplete' is technically accurate but deeply misleading, because the models build a high-dimensional internal state via attention mechanisms befo…

A data analyst has developed five AI prompt templates that consistently deliver production-ready output for data workflows, including data cleaning, SQL optimization, root cause analysis, executive dashboard design, and …

Nylas has launched a Notetaker API that allows developers to programmatically join, record, and transcribe meetings from Zoom, Google Meet, and Microsoft Teams. The API supports both grant-scoped notetakers tied to a con…

A developer built a real-time desktop AI copilot for online calls that displays hints within seconds. The system captures system audio (not microphone) using ScreenCaptureKit on macOS and WASAPI loopback on Windows, stre…

Neowin published an editorial on Jun 21, 2026 listing feature requests for Google's Gemini assistant, including message deletion and UI navigation improvements. The piece traces Gemini's lineage from Bard and earlier mod…

Developers at Zone01 Kisumu completed a week-long Bitcoin programming bootcamp culminating in a 24-hour hackathon where 17 teams built real-world solutions using Bitcoin, the Lightning Network, and LND. One team develope…

A developer explains that large language model function calling is actually trained, constrained text generation, not real code execution, and walks through the end-to-end cycle of tool calls in agentic AI systems.

Vexyn launches a suite of browser-only privacy tools that process data locally using WebGPU AI, with no file uploads, signup, or tracking. The tools include JSON formatting, Base64 encoding, QR generation, PDF merging, i…

Researchers introduced GEPA, a prompt optimizer that uses an LLM to read execution traces and rewrite prompts, outperforming reinforcement learning by leveraging natural language reasoning about failures. The method main…

The U.S. Golf Association launched Rules AI, a chatbot-like tool using generative AI to help golfers interpret the sport's complex rules, at the 126th U.S. Open. Built with Deloitte and trained on over 25,000 real questi…

TestGrid outlines key QA techniques for testing voice-enabled AI chatbots, emphasizing challenges such as speech recognition accuracy, background noise, accents, and device behavior. The company recommends assessing tran…