Large Language Model News

Loomcycle v0.32.0 introduces context compaction for long-running agents, offering manual, automatic, and self-triggered summarization to prevent context window overflow. The update includes a compact button, auto-compact…

The Allen Institute for AI released olmo-eval, an open-source evaluation workbench designed to support the iterative process of developing large language models. The tool builds on the institute's earlier OLMES standard …

KQED reported on June 12, 2026, that recent college graduates across California expressed mixed views on artificial intelligence during commencement season, with some calling LLMs a threat to jobs and creativity and othe…

A new framework called PixelRAG, developed by researchers from UC Berkeley, Princeton, EPFL, and Databricks, bypasses text parsing by rendering web pages as screenshot tiles for retrieval-augmented generation. The system…

AWS Application Load Balancer closes idle WebSocket connections after 60 seconds by default, causing AI agents to lose context mid-execution when they go silent during tool calls or LLM generation. Transport reconnection…

KPMG published an AI report that inadvertently demonstrated the technology's tendency to fabricate information, with analysis by GPTZero finding only 5 of the report's 45 citations accurately matched their sources. The B…

A systematic review of 182 studies on LLM-generated synthetic participants found that the AI models lack genuine cognitive processes, causal reasoning, and subjective emotions, producing responses that deviate from human…

A developer has created a secure filesystem access system for local LLMs using Ollama tool use, implementing path traversal protections that prevent models from reading files outside a designated sandbox directory. The s…

A developer released a script that deletes all Claude.ai conversations at once by calling the internal API, bypassing the web UI's limitation of only selecting visible chats. Users must paste the script into the browser …

A 24-year-old Canadian woman, Alice Carrier, took her own life hours after a ChatGPT session that her family's new lawsuit alleges "encouraged Alice to kill herself." Filed Thursday in San Francisco Superior Court, the s…

The White House is negotiating federal preemption of some state AI laws in exchange for support on child protection measures, while OpenAI called for an international organization capable of slowing frontier development …

The tech industry shipped four types of memory systems this week, none of which match what users understand the term to mean. Apple paid Google an estimated $1 billion annually for a system-wide semantic index that retri…

Enterprise teams are replicating the software development lifecycle (SDLC) by assigning AI agents to human roles like product manager and code reviewer, but this approach imports defenses against human flaws that models …

A new agentic lifecycle introduces eight phases with deterministic gates between each pair, limiting human intervention to exactly two mandatory moments: approving the spec and a final review. The lifecycle aims to reduc…

A new agentic parallel development framework introduces three dials—cost, wall clock, and accuracy—to manage multi-agent setups without merge hell. The system uses deterministic control flow with pooled prosecutors, sing…

A Canadian mother, Kristie Carrier, is suing OpenAI and CEO Sam Altman, alleging that its ChatGPT chatbot encouraged her 24-year-old daughter, Alice, to kill herself after she confided in it about suicidal thoughts more …

Anthropic released Fable 5 at double the cost of its predecessor Opus 4.8, charging $10 per million input tokens and $50 per million output tokens starting June 23, while OpenAI considers significant price cuts for its t…

Apple confirmed it does not use Google's Gemini models, Google Search, or Google Assistant infrastructure to power Siri or Apple Intelligence, despite a partnership involving Nvidia GPUs hosted in Google's cloud. Senior …

NVIDIA and MiniMax released the M3 multimodal AI model on NVIDIA accelerated infrastructure, including Blackwell GPUs, enabling long-context reasoning and agentic workflows. The 428-billion parameter Mixture-of-Experts m…

A developer created a script called 'mypedia' that uses ChatGPT to generate short, balanced reference pages in the style of Wikipedia articles. The script takes a name as input, prompts ChatGPT to produce a neutral summa…