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 warns that internal Slack runbook bots sending incident prompts to OpenAI's servers can expose sensitive operational data, leading to compliance failures during SOC 2 audits. The post provides a compliance ch…
GitHits launched its public beta 0.9, a CLI and local MCP server that gives coding agents access to open-source code behind project dependencies. The tool indexes open-source code and documentation, allowing agents to se…
A developer is building production-style AI engineering systems that integrate AI as a component within full backend architectures, rather than simple API wrappers. Projects include an AI personal assistant, a multi-agen…
Cursor benchmarked its Composer 2.5 and Composer 2.5 Fast models across 11 engineering skills, finding that the fast model scored higher, ran 32% quicker, and costs the same. The fast model achieved an average with-skill…
A developer built a resumable overnight build loop using an autonomous coding agent that starts each session with zero memory. The system externalizes state through granular checkpoints, decision logging, and guardrails …
Anthropic faces a proposed class-action lawsuit alleging its Claude Max 5x and Max 20x subscriptions deliver far less usage than advertised. Plaintiff Karl Kahn claims the $200 Max 20x plan provides only six to eight tim…
LLMs are shifting from vibe-coding to specification-based programming, where human input is via specifications that generate code. This approach enables controlled, repeatable software creation, with features and bugs ma…
XAI shipped the Grok Build Agent Dashboard on June 15, 2026, enabling developers to run up to eight parallel coding agents from a single terminal interface. The dashboard automatically sorts sessions by state, groups sub…
Engineers achieved up to 67% cost savings and 2.7x better goodput by using Prefill-Decode disaggregation with Ray and vLLM on AMD MI325X GPUs, separating prefill and decode phases onto dedicated hardware to eliminate mut…
A new open-source project called SAMF introduces machine-readable validation contracts for LLMs, translating MoSCoW prioritization into deterministic guardrails for multi-agent loops. The tool aims to eliminate brittle m…
Search Engine Journal reports that businesses can increase their odds of being recommended by AI-driven local search tools such as Gemini, Claude, ChatGPT, and Perplexity by adopting a continuous review-generation strate…
Anthropic suspended access to its Fable 5 and Mythos 5 models for all customers after the US government issued an export control directive citing a potential jailbreak that could identify software vulnerabilities. The mo…
An audit by NewsGuard found that Mistral AI's chatbot, Le Chat, repeated state-sponsored disinformation from Russia, China, and Iran 50% of the time in English and 56.6% in French, marking a significant increase from a 3…
A developer released FlashQwen, a from-scratch CUDA inference engine for Qwen3-8B, built with C++ and CUDA. The project is hosted on GitHub and aims to provide efficient inference for the Qwen3 language model.
A developer created a method to run multiple isolated Claude Desktop instances on macOS, each with a color-coded Dock launcher that focuses existing windows instead of spawning duplicates. The solution uses Electron's --…
Developer Anil Madhavapeddy released ocaml-deepseek, an OCaml library that integrates DeepSeek's open-weight LLM directly into applications via a native inference engine, enabling local, dependency-free AI agents. The pa…
Researchers introduced CacheWise, a KVCache management layer for LLM coding agents, reducing evictions by 2-2.6x and improving session completion time by up to 3.5x in vLLM, according to a June 2026 arXiv paper.
Tangram, a system described in an arXiv paper submitted on 15 June 2026, hides GPU heterogeneity to allow existing LLM parallelizers to operate on heterogeneous clusters, reporting up to 2.3x higher training throughput v…
Researchers introduced LOGOS, a generative foundation model for the natural sciences that encodes scientific objects and spatial interactions as token sequences. Trained at 1B, 3B, and 8B parameters, LOGOS matches or out…
A developer ran tests comparing AI companion behavior across models and relationship architectures, finding that a cheaper model with relationship architecture behaved more like a person than a stronger model with a stat…