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 introduced JetSpec, a speculative decoding method that trains a causal parallel draft head on a frozen target model to draft entire speculative trees in one pass while preserving autoregressive factorization.…
AI in Context released a video detailing the five-day crisis in November 2023 when OpenAI's board fired CEO Sam Altman, leading to employee revolt and his reinstatement. The video explores corporate governance and accoun…
Cursor has acquired Continue, an open-source coding assistant and alternative to GitHub Copilot, in a move that consolidates the AI developer tools market. The acquisition brings Continue's open-source platform under Cur…
Anthropic reported elevated error rates across multiple Claude AI models, triggering incident notifications for subscribers. Users can receive updates via email or SMS when incidents are created or resolved.
Researchers introduced Self-Harness, a new paradigm enabling LLM-based agents to iteratively improve their own operating harnesses without human intervention. In tests on Terminal-Bench-2.0, Self-Harness boosted pass rat…
Sakana AI launched Sakana Fugu, a multi-agent orchestration system that routes tasks across a pool of frontier LLMs via a single API. The system, available in two variants (Fugu and Fugu Ultra), outperforms individual mo…
Google DeepMind has made the Interactions API the default interface for Gemini models and agents, replacing the old generateContent API. The new API uses a simplified schema with typed steps instead of role-based structu…
Micron and Anthropic have signed an AI infrastructure supply agreement, under which Micron will provide memory and storage solutions to support Anthropic's AI model training and deployment. The deal underscores growing d…
Kilo announced Next-Edit, a feature powered by Inception's Mercury Edit 2 model that predicts code edits based on context, and is making it free for all users for one month. The diffusion LLM architecture enables paralle…
Amazon is testing its generative AI-powered assistant Alexa+ in India with Hindi support, inviting users to join a beta program by June 22. The company aims to tap the large Hindi-speaking market, though the rollout time…
Large language models are proving transformative for blind and low-vision users by enabling intent-driven, conversational interaction that bypasses the linear, burdensome nature of traditional screen readers. Real-world …
Abacus AI's ChatLLM platform offers a multi-model AI workspace that integrates leading models like GPT, Claude, Gemini, Grok, DeepSeek, and Qwen under a single subscription, along with AI agents, coding tools, image gene…
HPE's Spaceborne Computer 4, destined for a lunar lander as part of the Astrolab FLEX mission, will use a Kioxia BG-series M.2 SSD for storage to run LLMs on the moon. The compact, low-power system relies on a commercial…
A 2026 University of British Columbia study found 21.8% structural waste across 4.45 billion tokens in LLM context windows, proposing demand paging as a solution. The researchers argue that treating the context window as…
AWS and Vexcel developed a multimodal AI system that uses embeddings, LLM captioning, and vector search to make aerial imagery searchable via natural language, eliminating the need for per-feature model training. The sys…
GitHub Copilot for JetBrains IDEs now supports organization and enterprise agents from GitHub, allows queuing and steering messages in CLI sessions, introduces an agent debug logs summary view, and brings Claude as an ag…
A developer found that giving an AI coding agent a tool to check code health before commits achieved 100% compliance, compared to only 70-85% with a written rule. The tool, an MCP server entry point called code_health_sa…
Claire explains how to design AI agent loops in Claude Code and Codex, covering heartbeats, crons, hooks, and goal-based loops, and builds real automations including a daily PR-review loop and a weekly skills loop that s…
VLLM, a large-model inference serving framework, uses Python for control flow but pushes arithmetic into compiled C++ and CUDA kernels to avoid interpreter overhead. The Python/C++ boundary crossing incurs fixed costs fr…
A new analysis argues that claims about large language models having preferences, such as 'Claude prefers React,' are misleading because models lack preferences and instead reflect training data and context. Researchers …