Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Google's Gemini 3.5 Flash narrowly defeated Anthropic's Claude Opus 4.8 in a head-to-head benchmark, scoring 35.4 to 34.8 across four tasks. Gemini won three of four tasks by demonstrating superior instruction-following,…
A developer created Agent Swarm, a pipeline where nine specialized AI agents collaborate to code, test, and deploy software automatically. The system includes agents for orchestration, search, planning, contracts, fronte…
Chinese open-source model GLM-5.2, released by z.AI under an MIT license, is challenging U.S. AI leaders like Anthropic and OpenAI by offering competitive coding and agentic-workflow performance on a one-million-token co…
Google's GKE Labs introduced OpenRL, an open-source self-hosted API for post-training and fine-tuning large language models on Kubernetes clusters. The project decouples reinforcement learning infrastructure from AI rese…
Daikin Applied Americas uses Databricks Genie Code to build consistent data pipelines at scale, implementing a MECE skill framework to enforce architectural standards and accelerate pipeline development from days to minu…
A developer compared five vector databases—pgvector, Pinecone, Qdrant, Weaviate, and Milvus—for RAG applications, finding that the best choice depends on existing infrastructure, operational preferences, scale, and hybri…
Zep reports that placing agent memory in a trailing message instead of the system prompt can cut token costs by up to 2x by preserving prompt caching. The technique works on any provider and savings grow with conversatio…
A developer outlines a three-phase Claude Code workflow that shifts from 'Vibe Coding' to 'Agentic Engineering' by using structured prompts with explicit requirements, planning, and verification criteria to reduce wasted…
A researcher has developed a technique called 'ontological inversion' that allows large language models to capture the multifaceted nature of complex concepts like sorrowfuljoy, rather than overfitting to singular emotio…
Google's Gemini in Chrome now includes a 'Select from screen' tool that lets users add images to prompts, rolling out with Chrome 149. Separately, the Gemini 3.5 Flash model gains native computer use capabilities for dev…
A developer built a production-grade API for a 3B text-to-SQL model using FastAPI and Ollama, enabling natural language queries against SQLite databases at zero inference cost. The API, part of the de-swarm project, incl…
A developer argues that large language models do not possess understanding or mind-reading abilities, but instead predict the next most likely tokens based on patterns in user input. The post emphasizes that the real ski…
A developer named Maneshwar, building the git-lrc AI code reviewer, clarifies what an 'agent' truly means in AI, defining it as a system that independently completes tasks. The post outlines the core components of agents…
Google launched Gemini 3.5 Flash on May 19, 2026, its first production AI model capable of navigating computer interfaces and executing multi-step workflows, scoring 78.4% on the OSWorld-Verified benchmark. The model, pr…
Shopify's head of engineering Farhan Thawar confirmed the company scrapped its internal AI token leaderboard after employees engaged in 'tokenmaxxing'—burning tokens to climb rankings rather than for useful tasks. Shopif…
Strands Labs released AI Functions, a framework that lets developers write Python functions by describing their behavior in plain English, with runtime validation to ensure correctness. The framework uses Amazon Bedrock …
OpenAI and Broadcom unveiled their first custom AI chip, Jalapeno, which shows cost savings of roughly 50% compared to typical AI GPUs. The chip is being tested and will be integrated into data centers from Microsoft and…
Researchers from the UK and Germany identified three key drivers behind 'AI psychosis' in a study published in Nature's Digital Psychiatry and Neuroscience. The drivers—linguistic alignment, hyperpersonalization, and the…
Snowflake CEO Sridhar Ramaswamy reported that Zhipu AI's GLM-5.2 model nearly matches Anthropic's Claude Opus 4.7 in a benchmark of 103 coding tasks, while costing one-fifth as much per output token, though GLM-5.2 uses …
OpenAI unveiled Jalapeno, its in-house LLM inference accelerator chip developed with Broadcom, achieving tape-out in nine months. The chip is designed to accelerate inferencing for OpenAI's LLM stack powering ChatGPT, Co…