Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Researchers from Princeton and other institutions introduced CoALA (Cognitive Architectures for Language Agents), a framework that structures language agents with four memory systems: working, episodic, semantic, and pro…
Google's AI Overviews now appear in about one in five searches, reducing organic click-through rates to 8% when present. A new optimization strategy focuses on earning citations within these AI summaries rather than trad…
Researchers developed XBAR, an auditable latent crossbar memory fabric that enables model-to-model communication by writing directly into a frozen transformer's KV cache. The system achieves O(1) VRAM scaling with contex…
A developer describes the evolution from prompt engineering to context engineering and finally to building a 'harness'—a persistent structure of files, memory, and tool access that enables AI models to work effectively w…
Pi Coding Agent from Earendil Works is a minimal, extensible architecture for terminal-first AI coding workflows. It separates responsibilities into layers including provider abstraction, agent core, coding-specific logi…
Anthropic Managing Director of International Chris Ciauri said during a press conference in Seoul that the company expects to restore access to its frontier AI models, including Fable 5, within days after blocking them f…
Z.ai released GLM-5.2, an open-weight model that the author calls the best open-weight model available. The model introduces IndexShare, a cross-layer reuse trick for DeepSeek Sparse Attention that reduces the cost of 1M…
A developer reveals three structural cost leaks in LLM pipelines: uniform model routing, synchronous processing, and lack of caching. After addressing these issues, the same workload can cost far less without changing bu…
A developer has published a guide for setting up a local AI coding assistant in VS Code using Continue and Ollama, achieving tab autocomplete and code chat entirely on-device. The setup requires a GPU with 24GB+ VRAM for…
A developer has created a guide for building a private, local ChatGPT alternative using open-source LLMs like Qwen3 14B or Llama 3.1 8B, Docker, and a GPU with 12GB+ VRAM. The setup runs entirely offline, offers multi-us…
Openfusion, an open-source drop-in compound-model proxy, lets users point any OpenAI-compatible tool at it to fan out prompts to a panel of LLMs in parallel, then a judge model synthesizes a single answer. The project ai…
Ray Serve LLM, in partnership with Google Kubernetes Engine, announced major performance improvements achieving up to 4.4x higher throughput on prefill-heavy workloads and 24x higher on decode-heavy workloads through opt…
The insurance industry is undergoing a transformation driven by AI, with a shift toward contextual, conversational, and customer-first experiences. The Zebra's new chief AI officer, formerly of Google, emphasizes using l…
The UK Home Office is using two AI tools, Asylum Case Summarisation (ACS) and Asylum Policy Search (APS), both powered by GPT-4o, in asylum decision-making without informing applicants or publishing required impact asses…
OpenAI is testing a new subscription called ChatGPT for Science, aimed at scientific and academic use cases, according to reports from BleepingComputer and TugaTech. The subscription is linked to OpenAI's GPT-Rosalind mo…
JPMorgan Chase has blocked its Hong Kong staff from using Anthropic's Claude AI model due to licensing terms, following a similar move by Goldman Sachs. The restriction highlights how geopolitical tensions and US export …
A developer built opsx, a workflow tool that enforces a spec-first approach to AI-assisted coding, requiring a reviewed change document before any code is written. The tool separates management and governance planes, pro…
A new study from the Wharton School at UPenn introduces 'cognitive surrender,' where AI users accept AI-generated answers without independent verification. In experiments with 1,372 participants, 73% accepted wrong answe…
AI models used for online hate speech detection vary widely in how they identify and classify hate speech, according to a 2025 University of Pennsylvania study. The study found significant inconsistencies across systems …
Anthropic's AI chatbot Claude suffered a major outage on Thursday, with Downdetector recording over 2,000 user complaints. The company identified the cause and is working on a fix.