Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Claude Code lacks persistent memory across sessions, losing architectural decisions, bug fixes, and project context. Tools like Memarch, Hermes, and GBrain offer solutions for storage, context injection, and semantic rec…
OpenAI rolled out live job search and resume tailoring features to ChatGPT in 2025, allowing users to find job listings and customize applications through natural language prompts. The feature pulls live listings from ma…
Sakana AI's Fugu multi-model orchestration system costs 5x more and runs 4.5x slower than Anthropic's Claude Opus 4.8 while delivering similar results, raising questions about the practical value of multi-agent approache…
Ai2 released Tmax-27B on 23 June 2026, an open-weight terminal-agent model built on Qwen3.6-27B that scores 43% on Terminal Bench 2.0 and 69% on TB Lite. The dense 27B model outperforms the sparse 397B model on coding be…
Cursor developer Anysphere is reportedly training a 1.5 trillion parameter large language model from scratch using SpaceX compute infrastructure, signaling a shift from being a model wrapper to a frontier AI lab. The com…
Visa built conversational BI agents on Authorize.net payments data using ClickHouse Cloud and LibreChat, enabling natural language queries over 40 petabytes of data. The system reduced multi-day reporting to sub-second a…
ClickHouse announced the Managed ClickStack MCP Server for ClickHouse Cloud, enabling AI agents to access specialized observability tools for logs, metrics, and traces. The server provides higher-level semantic investiga…
Vercel's AI Gateway now offers GLM 5.2 Fast via Wafer, delivering 2x higher throughput than other providers in benchmarking tests. The model achieves over 170 tok/s for small contexts and over 200 tok/s for large context…
A developer implements a system where Claude Code writes code diffs and a separate Codex model reviews them before human approval, using cross-model disagreement to catch bugs that self-review misses. The hook-based work…
A developer argues that human attention, not agent capability, is the bottleneck in AI-assisted workflows. While agent capacity and compute costs are dropping, the fixed resource of human review time limits productivity.…
Engineers can now understand the transformer architecture behind large language models through a plain-English guide that explains how words become vectors, layers move points, and attention mechanisms allow context-awar…
A writer describes how to add an anti-slop pass to a blogging or artifact pipeline to remove AI-generated tells from drafts. The method uses an open-source catalog of AI phrases and structures, running the pass after the…
AMD introduces DP Attention and Two-Batch Overlap (TBO) optimizations for DeepSeek-V4 inference on MI355X GPUs, using a coordinated prefill scheduler called PrefillDelayer to reduce padding waste and phase mismatch in Mo…
Google at I/O 2026 integrated Firebase setup directly into Antigravity 2.0 and Android Studio, eliminating the need to leave code for the Firebase console. Agent Skills for Firebase expanded from web-only to mobile, enab…
Google's Agent Development Kit (ADK) and the Agent2Agent (A2A) protocol enable a cross-language multi-agent system where a Python agent using Gemini extracts contract terms and a Go agent validates compliance, solving th…
Leaked documents from Russia's Social Design Agency reveal plans to build a network of Wikipedia-style sites, media outlets, and fake think tanks to manipulate search engines and AI chatbots with Kremlin-friendly narrati…
In a series of four preregistered experiments involving nearly 19,000 conversations, AI systems consistently outperformed expert human persuaders—including professional canvassers and world champion debaters—even when hu…
OpenAI released Plant Talk, an open-source project that lets users converse with a houseplant via ChatGPT using a webcam, microphone, and the OpenAI API. The system can optionally integrate Arduino sensors for more accur…
Anthropic launched Claude Tag, a feature that lets users assign tasks to its AI assistant @Claude within Slack channels, where it can use memory, scheduled tasks, and tool access for ongoing work. The company reports tha…
A developer built a production-ready LangGraph ReAct agent that exposes an OpenAI-compatible API, supports multi-model switching via a gateway, and includes one-line tracing with Langfuse. The deployment uses a FastAPI r…