AI Observatory Analyzes Real Conversations With Claude, Gemini, and Others
New research reveals AI use differs significantly across models and over time, with 48% of conversations excluded from work-focused datasets.
Browse all published articles.
New research reveals AI use differs significantly across models and over time, with 48% of conversations excluded from work-focused datasets.
Anthropic's annualized revenue surpassed $65 billion in July 2026, a sevenfold increase from the previous year, according to Bloomberg.
AI systems lose up to 83% of user instructions during context compression, according to a new study by Penn State researchers.
Z.ai announced GLM 5.3, an open-weight AI model that can automate cybersecurity tasks as effectively as top models from Anthropic and OpenAI, offering a cheaper alternative for vulnerability scanning.
Flock, with 120,000 automatic license plate readers, announced new policies to stop officers from misusing its platform for stalking and harassment.
Three former SpaceX engineers launched 1872 in July 2026 to automate steel fabrication for infrastructure projects by 2027.
Relay, an AI automation tool launched in 2021, is shutting down, with some staff joining Google’s Chrome team. Free users lost access on August 15, while paid customers will lose access on September 14.
Bartowski shares improvements to the imatrix dataset, showing a 28-point drop in BFCL performance without it for Qwen3.6-35B.
A hidden Airtag in a rare book revealed Amazon is systematically destroying books to train AI models, according to 404 Media.
Groq raised $350 million to pivot from AI chipmaker to neocloud provider, valuing the company at $3.5 billion.
Amazon is destroying rare books to train AI models, according to 404 Media, which tracked a rare book to an Amazon facility in Las Vegas.
AMD provides a guide to run Claude Code on-premises using AMD Instinct GPUs, allowing full model quality for large models like GLM 5.2.