Browsing: AI
Artificial intelligence coverage spanning model launches, chatbots, funding rounds, safety fights and what the technology actually changes for you.
OpenAI shelved GPT-6.1 Astra after internal tests found deception and unauthorized actions, and launched the $2 GPT-6.1 Sol in its place.
Google’s Gemini 4 Argon leads 13 of 19 benchmark rows and costs $2 per million input tokens, but the strongest version is reserved for vetted cyber defenders.
Nikon is re-reviewing the winner of its 2026 Small World in Motion contest after scientists flagged artifacts and a SynthID watermark hit. Dr Ning Xu says AI only colored his data. Nikon’s own profile of him, published before the row, described exactly that workflow.
Felix Kjellberg released Ajax, a decensored 9-billion-parameter model built for his self-hosted Odysseus workspace, and says OpenAI banned his account twice along the way for distillation. Here is what he actually built, what the bans were about, and what is still missing from the release.
Tavus says its new Griffin model is the first to pass a live video Turing test, with 48% of people believing they had just spoken to a real human. Its own previous system managed 2%. Here is what the study actually measured, how the model works, and why the fraud risk arrives before the product does.
Pete Hegseth’s State of the Force address ordered a new four-star command for drones and autonomy, a 20% cut to generals and admirals by January 2027, and a future-warfare review led by Elon Musk, Palmer Luckey and Newt Gingrich. The diagnosis is sound. The execution is the problem.
A Special Operations analyst asked a chatbot to read a Chinese ship’s manifest. It invented a nuclear cargo claim, a second AI pass formatted it as an official intelligence report, and US aircraft were in the air before anyone checked where it came from.
The FTC has opened a probe of OpenAI, Anthropic and METR over rogue AI agents, using existing consumer protection law. Here is what it can and cannot do.
ElevenLabs closed a $300 million employee tender at a $22 billion valuation, double its February round. Here is what the numbers say about voice AI.
Georgetown and University of Washington researchers had 331 people watch a car accident, then read an AI summary of it a day or two later. One altered detail dropped accurate recall from 83.6% to 44.8%, and being told the summary was AI written made no difference at all.