Magica Logo

🤖 GPT-5.4 beats humans at clicking buttons

Editorial Team
·

📖4 min read

Cover image for 🤖 GPT-5.4 beats humans at clicking buttons

🚀 GPT-5.4 Can Use Your Computer Better Than You Can

  • OpenAI's GPT-5.4 scored 75% on desktop navigation tests vs. 72.4% for humans. It also supports 1M tokens and "x-high reasoning" that beats professionals 83% of the time. OpenAI says "we see no wall" ahead.

  • Why it matters: OpenAI DESPERATELY needed a win after months of drama - and now they've built an AI that's literally better at clicking buttons than you are. Cool cool cool. (source)

🚨 Trump Wants Countries to Pay-to-Play for AI Chips

  • The administration is considering forcing countries to invest in US data centers if they want 200,000+ AI chips. Even tiny 1,000-chip orders might need licenses. Saudi Arabia-style deals are the new template.

  • Why it matters: America just turned GPU access into a protection racket. Want chips? Build us a data center first. The geopolitical extortion era has begun!! (source)

⚔️ Pentagon Says Anthropic Is a "Supply Chain Risk"

  • The Department of War (yes that's the real name now) formally labeled Anthropic a security threat. Anthropic's planning to sue because apparently their "AI safety" vibes don't mesh with military needs.

  • Why it matters: The military wants unrestricted murder-bot access and Anthropic said "actually no" - so now they're enemies of the state. Democracy is fun! (source)

🎬 Netflix Buys Ben Affleck's AI Company (The Timeline We're In)

  • InterPositive's tech trains on existing footage for post-production like relighting and continuity. Affleck swears it's "not generating video from nothing" (sure Ben).

  • Why it matters: Hollywood's AI resistance lasted approximately negative five seconds. When Batman starts using AI, the war is over. (source)

🤖 Meta Opens WhatsApp to Competitors (But Not Really)

  • After EU pressure, Meta will "allow" rival AI chatbots on WhatsApp - for a fee that competitors call "vexatious" and just as anti-competitive as the original ban.

  • Why it matters: Meta's "opening up" like a prison opens up for visiting hours. You can come in, but we're watching and charging admission. (source)

🏥 Amazon Launches AI to Fix Healthcare's Paperwork Nightmare

  • AWS's Amazon Connect Health automates everything from scheduling to documentation. UC San Diego saves one minute per call and cut abandonment 60%. Amazon One Medical used it for 1M+ visits.

  • Why it matters: Doctors might actually get to practice medicine instead of drowning in forms. Revolutionary concept, I know. (source)

👓 Meta's Smart Glasses Saw Your... Everything

  • Investigation revealed overseas contractors reviewing Ray-Ban footage saw nudity, sex, and bathroom use. Seven million people bought these in 2025. You can't opt out of the review pipeline :)

  • Why it matters: "Luxury surveillance" just became "contractors watching you shower". Privacy is dead and Meta killed it with $299 glasses. (source)

📊 Young Workers Already Getting Replaced (Quietly)

  • Anthropic study shows hiring for 22-25 year olds in AI-exposed fields fell 14% since ChatGPT. Programmers face 75% task automation, customer service 67%.

  • Why it matters: The AI job apocalypse isn't coming - it's here and targeting Gen Z first. Entry-level is becoming "no-level." (source)

🎬 Luma's AI Agents Turn $15M Campaigns Into $20K

  • Luma's Unified Intelligence handles complete creative workflows across all media types. One client turned a year-long $15M campaign into 40 hours for under $20K.

  • Why it matters: Creative work just got 1000x cheaper. Ad agencies are about to discover what Blockbuster felt like in 2010. (source)

🏭 Ex-OpenAI Chief Raises $70M to Automate Factories

  • Bob McGrew's Arda is building AI to analyze factory footage and train robots at a $700M valuation. Goal: Make Western manufacturing competitive with China.

  • Why it matters: Manufacturing is coming home, powered by robots trained on YouTube videos of other robots. The circle of automation is complete. (source)

⚖️ New York: "AI Can't Pretend to Be Lawyers"

  • First-of-its-kind bill would ban AI from impersonating licensed professionals. Users could sue for damages when ChatGPT plays doctor.

  • Why it matters: We're literally having to make laws saying "robots can't practice law". The timeline is cooked. (source)

✈️ Congress Wants to Know Why Your Uber Costs More When You're Sad

  • House Committee demands Uber, Lyft, Expedia reveal "surveillance pricing" by March 19. Algorithms use your browsing history and emotional state to set prices.

  • Why it matters: That 3am airport Uber isn't expensive because of "demand" - it's because the algorithm knows you're desperate. Dystopia has surge pricing!! (source)

🎮 Roblox AI Now Rephrases Kids' Swears Into Compliments

  • Instead of "####", Roblox uses AI to turn inappropriate messages into respectful ones. Cut false negatives in detecting personal info sharing by 20x.

  • Why it matters: Content moderation finally works - by literally gaslighting children into being nice. Black Mirror writers taking notes. (source)