Liquid AI says its experimental 280M-parameter drafter speeds LFM2.5-VL-3B decoding, with support for llama.cpp, MLX-VLM and SGLang.

xAI says SpaceXAI is using Grok Bot to scale customer support, but has not released deployment details or performance metrics.

ThorstenMeyerAI.com has published the original analysis on AI Halloween displays, explaining how to tell AI features apart from ordinary automation and rec

OpenAI introduced MentalHealthBench to evaluate AI responses to mental health conversations. Independent review of its design and results is still pending.

Four AI models ran the same company through its worst week, every decision versioned. Only two closed the deal they’d earned — and the gap is replayable.

Thorsten Meyer outlines 24 uses for Jev, including three live publishing checks, and argues teams should measure heuristic failures before deployment.

Thorsten Meyer’s September 29 model comparison puts Opus 5.5 on building tasks and GPT-6.1 Sol on review, with cost and speed trade-offs.

A source-based comparison of seven named mini PC configurations, their memory and storage tradeoffs, and what buyers should verify for local AI.