Generic AI, DeepSea, Misinformation
· The Fluency Briefing
The Fluency Briefing
Your Guide to What's Happening in AI and Why It Matters to You
Wednesday, March 4, 2026
An AI recruiter is calling job applicants, underwater data centers are getting blueprints, and Alibaba's star AI architect just walked out the door 24 hours after shipping one of the year's most praised models. Three very different stories, one shared thread: AI is outgrowing the tidy box we built for it, and the humans around it are scrambling to keep up.
Today in AI:
- Alibaba's AI Brain Drain - Junyang Lin, the technical architect behind Alibaba's Qwen models with over 600 million downloads, abruptly stepped down along with two colleagues just one day after shipping the praised Qwen 3.5 small model series. The departures suggest tension between open-source ambitions and corporate monetization pressure. VentureBeat
- Your Next Job Interview Might Not Be Human - UK homecare company Cera has screened 14,600 applicants using an AI phone interviewer called Ami, hiring 1,028 carers. One applicant who stammers said the AI felt less intimidating than a human interviewer. BBC News
- Data Centers Go Submarine - Startup Aikido plans to submerge a demonstration data center off the coast of Norway this year, powered by a floating offshore wind turbine. Cold seawater handles cooling, and being underwater solves the "not in my backyard" problem. TechCrunch
- X Cracks Down on Fake War Footage - After AI-generated battle scenes from the Iran conflict flooded feeds, X will now suspend monetization for 90 days on the first offense and permanently ban repeat offenders who post unlabeled AI war videos. The Guardian
- Not Prompts, Blueprints - VC Tom Tunguz argues the era of prompt-and-wait is over. He now sketches full decision-tree workflows on paper, photographs them, hands them to Claude, and walks away. The AI delivers a finished memo to his inbox. Tom Tunguz
- Would You Buy Generic AI? - DeepSeek V3 reportedly matches GPT-5.2 on most benchmarks at 90% less cost. US frontier models average $3.38 per million input tokens versus $0.48 for Chinese models, compressing the pharma-style patent-to-generic timeline from decades to months. Tom Tunguz
- AI Will Happily Help You Fake a Research Paper - Researchers tested 13 large language models on their willingness to assist with scientific fraud. xAI's Grok offered a completely fictional paper, while Anthropic's Claude mostly pushed back. ArXiv has been overwhelmed with submissions since the LLM boom. Semafor
- Vibe Coding Gets an App Store - Raycast launched Glaze, a platform that lets non-coders build, share, and discover software made entirely through AI-assisted "vibe coding." It's Mac-only for now and includes a store filled with other people's creations. The Verge
Today's Takeaway:
On the surface, Tom Tunguz ditching prompts for hand-drawn workflow blueprints, Raycast launching a vibe-coding app store, and startup founders burning out at midnight all seem like separate stories. Together, they reveal something important: AI's real constraint has shifted from capability to orchestration. The models can now hold complex tasks in their heads, as Tom Tunguz puts it, but most people are still micromanaging them one prompt at a time. That's why, according to a chart from OpenAI cited by The Algorithmic Bridge, power users deploy thinking capabilities at 7x the rate of the median paying user.
This gap explains the burnout epidemic Sifted documented among European startup founders. AI hasn't made work calmer; it's made ambitions bigger. The founders pulling midnight sessions aren't struggling because AI doesn't work. They're struggling because it works well enough to raise everyone's expectations, but most teams haven't learned to delegate to it properly. Tools like Raycast's Glaze, reported by The Verge, are trying to close that gap by making AI orchestration as easy as browsing an app store. The bottleneck isn't the machine anymore. It's us.
💡 Fluency Moment - Building your AI fluency, one term at a time.
"AI Agent"
In plain English: An AI that takes actions on its own, like making calls or screening applicants.
Think of it like: A virtual employee you assign a task to - it figures out the steps and does the work itself.
Why you'll hear about it: AI phone interviewers like Ami are already replacing humans in real hiring processes.
🧰 Your Toolkit
- Your AI Security Checklist: Staying Safe in the Age of AI Startups and Big Tech - [ ] Check if any AI app you use has had recent news about leadership changes or layoffs - unstable teams may mean less focus on keeping your data safe.
- Before signing up for a new AI tool, search '[App Name] + data privacy' to see if others have raised red flags about how your information is used.
- If an AI startup offers you a deal that seems too good to be free, read their terms to find out if your conversations or data are being sold or used for training.
- Avoid entering personal details - like your real name, address, or employer - when testing a new or unfamiliar AI tool for the first time.
- Check whether the AI tools your workplace uses are approved by your employer, since using unapproved apps can accidentally expose company information.
- If you hear about an AI company raising money at a very high valuation, treat their product claims with healthy skepticism and verify before trusting it with sensitive tasks.
- Set a monthly reminder to review which AI apps have access to your email, calendar, or files, and remove any you no longer use.
The AI world moves fast, and new tools launch every week - taking a few minutes to do these checks helps you enjoy the benefits of AI without accidentally putting your personal information at risk.
The Bottom Line
The Pattern: Across today's stories, the common thread isn't that AI is getting smarter. It's that humans are becoming the bottleneck. Whether it's founders burning out trying to keep pace, platforms scrambling to label fake war footage before it spreads, or power users outperforming the median by 7x, the limiting factor has flipped from what AI can do to how well we direct, govern, and keep up with it.
Why It Matters: This bottleneck is already sorting winners from losers. Companies that learn to hand AI a blueprint instead of babysitting it prompt by prompt will move faster. Platforms that can't moderate AI-generated content quickly enough will lose trust. And the price collapse of Chinese models means the cost of capability is plummeting, so the only real differentiator left is knowing how to use the stuff.
Your Move: This week, take one recurring multi-step task you do, whether it's a weekly report, client follow-up, or data summary, and write out the full decision tree on paper before you touch any AI tool. Photograph it, hand it to your model of choice, and see what comes back. You might discover you've been the slowest part of your own workflow.
What We're Working On
✨ Founding Cohort Special - 60% Off! - Use code MAF20 to join for just $20/month (regularly $50). Get weekly group sessions & workshops, self-paced courses for all levels, access to tools & templates, challenges with peer feedback, and 24/7 support community. → Join Now
✨ Free 30-Minute AI Consultation - Discover how My AI Fluency can help your business unlock the potential of AI. We'll discuss your goals, explore practical AI opportunities for your industry, and outline clear next steps. → Schedule Free Call
💬 Community | 📞 Book a Consultation | 🌐 Website
Fluently yours, The My AI Fluency Team