Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
A developer replaced brittle regex and CSS selectors with a local LLM to extract product data from messy legacy HTML, achieving reliable results at a fraction of the cost of cloud-based models. The approach uses Ollama's…
Security researchers chained multiple large language model and web application vulnerabilities during a red team exercise to escalate from a low-privileged account to full admin takeover. The attack exploited insecure ou…
A developer has mapped the four most common LLM sampling parameters—temperature, top-p, top-k, and min-p—to their concrete effects on output distributions, providing a practical guide for production deployment without re…
AI Verdict launches a browser extension that queries ChatGPT, Claude, Gemini, and Perplexity simultaneously from a single interface, eliminating the need to switch between tabs. The tool displays all four AI responses si…
A developer spent three hours debugging a production LLM memory system bug where a `rollback` method wiped out entire conversation histories instead of just undoing erroneous operations. The root cause was a code refacto…
A developer has proposed replacing equality-based assertions with invariant checks for testing AI agents, arguing that non-deterministic outputs make traditional tests flaky or fake. The approach uses structural and fact…
An investigation by Investigate Europe and the Guardian found that major AI chatbots, including Meta AI, Google's Gemini, and xAI's Grok, recommended unlicensed offshore casinos and provided advice on evading gambling sa…
Researchers have developed Doc-to-Atom (Doc2Atom), a compositional parametric memory framework that decomposes documents into typed knowledge atoms compiled into independent micro-LoRA adapters. The system uses a query r…
Frontier AI models have consistently improved in intelligence over time, with Anthropic's Fable 5 (Mythos 5) now demonstrating gamechanging performance for software engineering tasks. Open-weight models trail the frontie…
Claude Opus 4.8's system card shows three main construct validity concerns in its alignment assessment: chain-of-thought monitoring misses reasoning that never surfaces in text, evaluation awareness is underestimated, an…
Naver Cloud presented a sovereign defense AI strategy at the Korea Defense Industry Development Expo on June 10, unveiling its HyperCLOVA X Omnimodal model that integrates text, voice, video and maps into a single operat…
A federal agent testified Thursday that Jonathan Rinderknecht, the man on trial for allegedly starting the deadly Palisades Fire, used ChatGPT to search for the home address of DoorDash CEO Tony Xu with the stated intent…
Researchers have developed a method to extract classification answers from large language models (LLMs) by reading their internal hidden states instead of generating text, achieving faster and cheaper results. The techni…
An open-source project called Guard Skills has been released as a code quality alternative designed to catch systematic failure modes in AI-generated code, such as hallucinated APIs, mock abuse, and premature abstraction…
A developer has identified critical security vulnerabilities in Retrieval-Augmented Generation (RAG) pipelines, warning that connecting LLMs to live data sources introduces risks including data poisoning, indirect prompt…
A developer has identified five JavaScript configuration failures that cause AI automation scripts written for Claude and OpenAI tools to break when run locally, despite the code being logically correct. The most common …
A new study published in the Journal of Writing Research found that artificial intelligence language models can reliably rewrite business documents only when given highly specific instructions, and without precise guidel…
Ray Serve LLM and vLLM on AMD MI325X achieve up to 67% cost savings by disaggregating prefill and decode phases in LLM serving, separating them onto dedicated GPUs to eliminate interference and improve goodput by up to 2…
MiniMax released the open-weights MiniMax M3 model on Modular Cloud, featuring a new Sparse Attention operation that achieves up to 15.6x speedup on decode while maintaining a 1 million token context window. The model is…
Alibaba's 1.7B parameter Qwen3-TTS voice cloning model was fine-tuned using Fully Sharded Data Parallel (FSDP) with PyTorch and Ray, demonstrating memory-efficient distributed training across 4 GPUs. The implementation a…