Mistral has released Mistral Large 4, a 1-trillion-parameter open-weight model with 49 billion active parameters, as a preview, with weights due by the end of October. We cover the specs, the sovereign AI pitch, the cyber defence claims, independent rankings, price per task, self-hosting memory and the licence question.
TikTok has unveiled Shopping Assistant, a conversational AI agent, and Buy Direct, a one-click checkout from the For You feed built with Shopify, Salesforce, Shoplazza and Stripe. We explain how it differs from TikTok Shop, the numbers behind it, the agentic commerce race, and what brands should ask first.
Instinct, the $10 billion personal AI agent, can now join group chats, even with friends who have never signed up. We explain its two-agent permission model, what its terms say about non-users and group purchases, how it compares with ChatGPT and Meta AI, and practical rules for families and teams.
Reflection AI’s first model, Beam, is a 501-billion-parameter mixture-of-experts model with 23 billion active parameters and Apache 2.0 weights due this month. We count its wins and losses against seven open rivals in Reflection’s own table, unpack the 3-4x compute claim, and work out what it takes to run.
OpenAI will add an invisible textGrain watermark to ChatGPT and Codex text in the EU to meet the AI Act. We break down how it works, OpenAI’s own detection and quality numbers, why the detector is gated, how it compares with Claude’s worldwide watermark, and what UK and EU organisations should do.
US universities from Yale to Cornell are moving away from AI detection software even as AI cheating grows. We look at a professor’s hidden-prompt trap that caught 60 of 350 students, the false-positive arithmetic, what a 1,057-faculty survey says, how assessment is being redesigned, and the lessons for UK employers.
Four TechCrunch Disrupt 2026 sessions, from Together AI, CapitalG and Pathway to Oumi, Nvidia and Ricursive Intelligence, tackle how founders choose between frontier APIs and open-weight models. We explain what open and closed really mean, what Menlo Ventures’ data and current list prices show, and give a five-question framework for deciding what to rent, customise or own.
Gemini Notebook has replaced its per-feature daily caps with one compute-based limit that refreshes every five hours up to a weekly ceiling, shows an estimated cost before you generate, and adds a Generate later queue for Video Overviews and Slide Decks. We explain how the new limits work, what each Google AI plan gets, and how to make your allowance go further.
Benjamin Riley argues in The Verge that our minds are not equipped to handle AI, calling it a ‘cognitive hot dog’. We explain his case against the brain-as-computer model, what cognitive offloading means, the evidence from 26,811 Chinese pupils and 3.2 million maths problems, the Norway, New York and Los Angeles school restrictions, where the essay overreaches, and how businesses can protect their people’s thinking.
In a virtual-reality office study presented at NordiCHI ’26, 189 workers split real money with AI assistants and gave the female-presenting agent, Johanna, 10.25% less than the identical male-presenting Johan. We explain the University of Limerick and Zurich research, how the gap compares with UK pay data, the say-do gap behind it, earlier evidence on gendered AI, and how businesses should design and test AI personas.
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 )
