In a simulation contest using Kerbal Space Program, ChatGPT (and similar LLMs) piloted a spacecraft based solely on text-based state and goals. It placed second overall, showing impressive agility with minimal prompt engineering and no specialized training. While hallucination risks remain, this milestone hints at a future where off-the-shelf LLMs could autonomously navigate spacecraft.
Meta is piloting a “cloud processing” feature that requests users’ permission to upload unseen photos from their camera roll to Facebook for AI-driven collages and themes. While Meta says the photos aren’t currently used for training its AI, its privacy terms don’t exclude that possibility. Users can opt out, but the move raises deeper concerns about future data usage.
The UK will be the first country to launch a national AI system to detect early signs of healthcare failures. Health Secretary Wes Streeting announced it as part of Labour’s reform agenda. The AI will analyze NHS data to flag issues before they escalate, aiming to restore public trust and improve patient safety across the system.
A study by Carnegie Mellon and Salesforce shows AI agents succeed in only 30–35% of multi-step office tasks, failing 70% of the time. Gartner predicts 40% of such projects will be abandoned by 2027 due to high costs and low value. Many vendors falsely market non-agentic tools as agents, fueling hype over real capabilities.
