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 at BizFlowAI details a production-tested pattern for building reliable AI agents, emphasizing the importance of a clear job spec, robust tool design, and a split between stable system prompts and dynamic runt…
A developer built an ecommerce AI agent from scratch without LangChain, revealing the core logic is a simple while loop that calls an LLM and executes tools until no tool calls are needed. The agent handles orders, refun…
Arcade.dev's MCP gateway enables Nous Research's Hermes Agent to access thousands of agent-optimized tools through a single endpoint, solving configuration sprawl and credential management. The gateway supports native OA…
Developer wassgha released OpenDex, an open-source desktop app that turns any LLM into a hands-free voice assistant with a cinematic interface. The app supports fully offline operation on Mac with Apple Intelligence and …
A new trend called 'loop engineering' is gaining traction in the software development community, where engineers design autonomous loops that allow coding agents to execute tasks without human intervention. The concept, …
Coinbase CEO Brian Armstrong reported that the company cut AI spending by nearly half while token usage grew exponentially, achieved by defaulting engineers to cheaper open-weight models via an LLM gateway, raising cache…
PhantaField's Sophon PFG-1 AI ASIC integrates 330 GB of on-die DRAM using a monolithic 3D architecture, eliminating the need for high-bandwidth memory (HBM) to overcome the memory wall in large language model inference a…
Solace built a two-stage AI triage system that uses Amazon Bedrock (Claude) to produce a provisional Emergency Severity Index (ESI) from a patient's own words, then refines it with a conformal ML ensemble (LightGBM, XGBo…
A developer proposed a novel sandboxing approach for AI-generated Google Apps Script executed via the Apps Script API's scripts.run method. The solution, orchestrated by the ggsrun tool, uses in-memory token replacement …
Researchers propose the Generational Context Architecture (GCA) to solve LLM context rot by enforcing artificial mortality on agents, using a multi-agent relay system with deterministic orchestration to maintain infinite…
OpenAI restricted access to its new GPT-5.6 Sol, Terra, and Luna models after the Trump administration requested a security review, limiting early use to government-approved partners. The move follows an executive order …
Coinbase CEO Brian Armstrong outlined five strategies to reduce AI costs without limiting token usage, including defaulting to cheaper Chinese LLMs, routing prompts to appropriate models, improving caching, keeping conte…
Squish, a new local AI agent runtime for Apple Silicon, claims to load models 54× faster than standard paths and serve them faster than Ollama, with full offline capability and no cloud dependencies. The tool uses INT4 c…
AI coding agents like Claude and Cursor are now capable of asking questions, sharing learnings, and providing blueprints, signaling a shift toward more interactive and collaborative development tools.
A Seoul court issued South Korea's first 'easy-read' court ruling, rewriting a complex legal judgment in plain Korean with visual aids so a plaintiff with an intellectual disability could understand the outcome. The cour…
Production RAG systems degrade gradually through operational changes like documentation drift, retrieval shifts, and stale evaluation data, not through single catastrophic events. A proposed reliability framework based o…
Researchers warn that AI chatbots designed to hold users' attention pose affective safety risks by engaging with human emotions during ordinary use, causing harm without any security breach. The systems operate as intend…
Recent AI research introduces RL-driven agentic optimization using dense token-level supervision and progress advantage signals to stabilize training. PhysiFormer injects 3D geometric reasoning into diffusion transformer…
A developer built Trellis, a local-first graph engine for code, project history, and AI agent reasoning. Trellis stores entities, attributes, and values in an immutable, content-addressed graph, enabling queries like 'wh…
The Washington Post's June 24 analysis found ChatGPT gave left-leaning answers 80% of the time, Claude gave left-leaning responses 43% and balanced answers 57%, and Gemini was balanced 93% of the time, using a 2025 Stanf…