Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Developers can deploy Llama 3.1 8B behind vLLM's OpenAI-compatible API on a rented GPU instance, achieving thousands of output tokens per second through continuous batching. The tutorial covers installing vLLM, authentic…
Google posted $109.9 billion in Q1 2026 revenue, its fastest growth since 2022, and still controls 90% of search, but talent defections, rising competitors like ChatGPT and Bing, antitrust remedies, and its own AI transf…
The era of subsidized AI API usage is ending as platform providers like OpenAI and Anthropic face unsustainable losses, with OpenAI reporting a $38.53 billion net loss in 2025. Hardware costs, particularly for DRAM and h…
Anthropic launched Claude Tag, a virtual employee chatbot that operates within Slack to autonomously complete tasks for teams. The tool, designed for enterprise use, allows multiple employees to collaborate with a single…
Andrej Karpathy proposes a markdown-based system for information retrieval using agents, arguing it preserves context better than RAG. The approach aims to improve how AI systems access and organize knowledge.
Anthropic has introduced Claude Tag (@Claude) for Slack, an always-on AI assistant that can be summoned in channels to help with tasks and share information. Currently in research preview for Claude Enterprise and Team c…
Anthropic launched Claude Tag, an always-on AI teammate for Slack that learns company context from conversations and can autonomously perform tasks. The beta feature for Enterprise and Team customers provides persistent …
A wave of research shows that AI agents which manage their own context window using learned policies outperform those relying on fixed truncation rules. Approaches like AgentSwing and budget-aware reinforcement learning …
GitHub has made the redesigned terminal interface for Copilot CLI generally available, featuring a tabbed layout for browsing issues, pull requests, and gists, a guided tool configuration experience, and a cleaner, more …
Google launched Ask Ad Manager, a Gemini-powered AI assistant for publishers using Google Ad Manager, to help troubleshoot delivery issues, generate reports, and navigate the platform via conversational prompts. The beta…
Kilo launched Auto Efficient, a new AI model routing tier that dynamically selects the most cost-effective model for each task based on real-time session classification and benchmark data. The system uses KiloBench to ma…
Software developers will need to adopt Hemingway-like conciseness as AI coding tools such as Claude Code, Codex, and Copilot prefer Markdown-style instructions, shifting from traditional human-readable specifications.
Z.ai's open-weights GLM-5.2 model (744B MoE, 1M-token context) is available for free via web chat and a rate-limited API through OpenRouter, with no credit card required. For developers, the free API tier offers about 50…
A major outage at Anthropic on June 23, 2026, disrupted Claude AI services, highlighting the risks of relying on a single LLM provider. The incident, caused by demand outpacing infrastructure capacity, followed similar d…
SkyPilot launched SkyPilot Endpoints, a production-ready LLM inference system that deploys and manages inference across multiple Kubernetes clusters from a single YAML specification. The system handles placement, autosca…
SGLang achieved a 5x throughput improvement for serving DeepSeek-V4 on NVIDIA GB300 hardware since its Day-0 launch, reaching ~11,200 tok/s/GPU at 50 tok/s/user through kernel and runtime optimizations including MHC fusi…
Krea AI released Krea 2, a series of foundation models for image generation focused on creative exploration, featuring a diffusion transformer architecture and systems for prompt expansion and style referencing. The mode…
Meta suffered its most embarrassing outage ever when a bug allowed users to change any account's email by simply asking the Meta AI bot, leading to high-profile account takeovers. Engineers attributed the disaster to AI-…
Neo4j launched the Agent Memory Service (NAMS), a managed cloud service providing persistent, structured memory for LLM agents using a graph database with native vector indexes. The service offers short-term, long-term, …
Markdown files are a legitimate pattern for agent memory in single-agent, single-user scenarios, but the approach breaks at scale, under concurrent agents, and as facts change and errors compound. Companies like Manus, C…