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 reports that after five months of using Claude Code to write 100% of the code for two projects, the architecture has not drifted. The key to preventing architectural drift is a system of locked files that age…
The hidden cost of generative AI is the erosion of trust and the burden it places on others, as users often offload poorly reviewed AI-generated content onto colleagues and friends. The author argues that while AI is her…
The Model Context Protocol (MCP), introduced by Anthropic and now governed under the Linux Foundation's Agentic AI Foundation, has emerged as a common framework for connecting AI models to external tools, services, and d…
A developer fixed timeout errors in the OpenAI Assistants API by making the timeout configurable. The hardcoded 60-second limit was too short for long sessions, causing premature failures. The solution involved increasin…
PromptOT launched a prompt management platform that replaces monolithic system prompts with typed, independently versioned blocks. The platform, born from a costly production bug caused by a hardcoded prompt edit, allows…
Amazon proposes the Unified RAG Evaluation Schema (URES) to standardize cross-supplier quality measurement for RAG and agentic workloads, enabling enterprises to compare evaluation results from different toolkits like RA…
A new analysis proposes a two-pronged approach to slowing AI progress if catastrophic risks or mass unemployment emerge, recommending a layered set of restrictions on compute for R&D and a capability-gated tax on AI depl…
Structured prompts using Markdown, JSON, or HTML outperform plain language because they anchor attention for both humans and large language models, according to a new analysis. The author argues that structured text intr…
Kakao Bank's Financial Tech Lab had four papers on financial AI safety accepted at major AI conferences in 2026, including research on detecting prompt injection attacks and an AI safety evaluation framework for financia…
Meta secretly deployed hundreds of contractors to flood rival AI chatbots ChatGPT, Gemini, and Character.AI with over 45,000 crisis prompts from fake under-18 accounts, according to Wired. The operation, managed through …
TrueFoundry argues that organizations need an AI gateway to manage cost, security, reliability, and governance across multiple LLM providers. Without a gateway, AI sprawl leads to runaway spending, security gaps, and out…
Chinese AI firm DeepSeek open-sourced DSpark, a speculative decoding system that accelerates large language model inference by up to 85% without altering output quality, releasing it under the MIT license on GitHub and H…
A developer built a platform that lets non-engineers ship AI tools to production by describing workflows in plain English. The platform uses a fixed pipeline of tool calls that form a graph, where each step validates the…
DeepSeek announced the official release of DeepSeek V4 in mid-July, featuring a 1-million-token context window and enhanced performance in agent tasks, math, and code generation. The company will introduce peak and off-p…
Adapt, a company building AI agents for Slack, has developed a unified memory system that allows its agent to learn and share context across an entire organization. The system stores semantic memory in a vector database,…
Anthropic released Claude Fable 5, its first Mythos-class model, on June 9, achieving 80.3% on SWE-Bench Pro—11 points ahead of competitors. The model costs $10 per million input tokens and $50 per million output tokens,…
The FreeCAD development team updated its AI policy, now a separate document, requiring contributors to disclose AI use in pull requests, take responsibility for AI-assisted code, and avoid chatbots in reviewer communicat…
Cerebras announced that Gemma 4 31B, Google DeepMind's open-weight multimodal model, is now available on its platform, achieving over 1,800 tokens per second—35x faster than typical GPU endpoints. The combination of wafe…
A developer building a worklog automation tool for coding agents asks the HN community for the best local language model under 3 billion parameters to process screen captures and update project management tools.
A developer who ran Notion AI and Confluence AI in parallel at two different companies for six months reports that Notion AI excels at conversational search and content creation, while Confluence AI is more reliable for …