The FAA will launch SMART, an AI planning system from Air Space Intelligence, starting in the Washington, DC area this fall. The $875 million, 12-year contract actually covers two systems, FMDS and SMART, and was awarded in June. We explain how the software works, why the controller shortage makes it urgent, where it fits in the wider air traffic overhaul, and the risks around cloud dependency, cybersecurity and measurable results.
PrismML, a Caltech spin-out, has released Ternary Bonsai 2 27B, a compressed Qwen3.8 27B that fits in 5.9 GB and retains 98.2% of the original’s aggregate benchmark score. We break down where it keeps and loses capability, the memory arithmetic behind the 9x claim, speed and energy figures, the Apple talks, where early coverage disagrees with the company, and what on-device models mean for businesses.
A Cambridge researcher, Lucius Caviola, argues that AI consciousness could become as divisive as abortion or climate change. His survey of 1,202 US adults found 45% think future AI consciousness impossible and 23% possible, and a 3,455-person preprint shows views polarise as AI grows more human-like. We set out the evidence, the expert disagreement between Anil Seth and David Chalmers, the four forces pulling opinion apart, and what organisations deploying chatbots should do now.
Google Labs’ Dreambeans got its AIxploria card on 18 September 2026, eight days after Google made the app free for adult US users. We checked the card against Google’s two announcements, TechCrunch’s launch interview and the Play Store. Eight of ten claims match, from the 10 to 14 daily stories to Nano Banana 2 artwork, and the “#19 in Storytelling Generator” badge sits exactly in slot 19. The card’s claim that Google does not train on Dreambeans data does not appear in any announcement, and a footnote excludes Workspace accounts.
Apolosign’s digital family calendar got an AIxploria card on 18 September 2026, with a 4.5 rating from six votes and a “#13 in Robots and Devices” badge. We checked it against Apolosign’s own pages. The founding date, household count, sizes and no-subscription model all hold. The 27-inch prices mix two sources into a pair no shop charges, the rival Skylight is quoted $20 above Apolosign’s own figure, the 2 million units claim is unsourced, and the one AI feature, Gemini chat, is never mentioned. The badge shares #13 with Project DIGITS, which looks like a tie.
A fortnight of AI doomsday warnings, from Jacob Coxon’s resignation to Dario Amodei’s call to pace frontier AI, has not deterred investors: Polymarket put an Anthropic IPO by year-end at 87%, and OpenAI is discussing a round at $1.2 trillion or more. We set out the timeline, the bull case (Gerstner, Buyer), the bear case (Palihapitiya, Burry, Eisman), what the companies say, and the SEC risk-factor rule that will force the debate into a legally binding document.
Figure’s Helix 2.5 humanoid AI model got its AIxploria card on 18 September 2026, about 16 hours after Figure’s announcement. We checked it against Figure’s own post: 30 unseen Bay Area homes, 56% zero-shot success against 9% without Index pretraining, 237 of 420 trials (beds 67%, towels 62%, toys 40%). The research numbers hold. The card leaves out the scaling law and the 1.90% cap on any task in pretraining, tags an unbuyable model “Paid”, repeats an unsourced sub-$20,000 price, and its exact #4 badge ranks the model above Figure 03, the robot it runs on.
A 92-page public version of the News Plaintiffs’ summary judgment brief, filed on 17 September 2026, quotes Microsoft’s Director of Applied Science Brent Hecht on “an astonishing theft of unprecedented proportions”. In full, the sentence forecasts how millions of people would see AI training. We read the brief and Microsoft’s own memorandum: the five stages of copying alleged, the datasets and counts, the “horse trading” between Microsoft and OpenAI, Copilot click-through drops of 83 to 93 per cent, the Bloom filter behind the “accidental cover up” line, and Microsoft’s 8.2-million-log defence.
Concerns about an AI-induced apocalypse are at a fever pitch, and biological weapons are the favourite mechanism. But the researchers closest to the work say the risk is lower than the headlines suggest: information was never the bottleneck, autonomous labs do not exist, humans in the loop can simply refuse, and pathogens make poor weapons. This article covers the four events that built the alarm, what the Stanford and Arc Institute phage study really showed, what Anthropic’s September threat report does and does not establish, where experts agree the risk is genuine, and which safeguards would actually move the odds.
Alibaba released Qwen3.8-Omni-Flash on 18 September 2026: an omni-modal model built on the Qwen3.8-Flash-Next architecture that accepts text, images, audio and video into a 1 million token context window and returns text. The headline is price — more than 98% off an hour of audio input against the previous generation — but the mechanism is more interesting. Agentic perception lets the model decide what to watch and listen to, raising OmniVideoBench accuracy from 63.4 to 67.8 while cutting tokens per query by about 45.7%. This article covers the specification, the benchmark claims, the pricing arithmetic, the Apache-2.0 plugin suite and the limits.
Progressive Robot: Your Gateway to Comprehensive IT Solutions — Specializing in Web Development, Mobile App Development, and Expert IT Services.
© All Copyright by Progressiverobot.com
VAT Number ( 506152326 )
