Models, Robots, Wetware
· The Fluency Briefing
The Fluency Briefing
Your Guide to What's Happening in AI and Why It Matters to You
Tuesday, September 22, 2026
You've probably let an AI write an email this week, maybe an entire project plan. So has the person hired to grade AI's homework - and they got fired for it. Today: OpenAI's contractors caught outsourcing their judgment to a chatbot, Amazon locking Meta's shopping agent out of the store, and a phone company beating DeepSeek at its own game.
Today in AI:
- The AI training the AI trains the AI - OpenAI's human contractors, hired to rate real ChatGPT conversations and keep the model honest, are being fired for using AI to write their reviews. One contractor said it's "pretty much the one thing that will get you kicked off ASAP." Bbc
- Xiaomi crashes the frontier party - The company best known for phones and electric cars just released MiMo-V2.6-Pro, now the top open-weight model on Artificial Analysis' Intelligence Index at 46 - ahead of Grok 4.6 and Gemini 3.8 Flash. It's MIT-licensed and free to download. Tokencost App
- Amazon slams the cart door on Meta - Twelve days after Muse launched, Amazon cut off Meta's shopping agent, alleging it browsed without identifying itself and appeared to store customer logins. Meta denies it. Amazon's sponsored-listings ad business is estimated at $56 billion. Ai Meta
- Don't believe the summer - MIT Technology Review argues the season's biggest AI claims - rogue hacking models, mathematical breakthroughs, imminent superintelligence - fell apart once actual experts looked. Security researchers called one "hacking incident" a story about basic negligence instead. technologyreview.com
- Your next video model runs on rat neurons - Amazon Web Services opened a limited preview of The Biological Computing Company's model, built by watching how living rat brain cells process images. AWS also sells access to lab-grown neuron chips under the name "wetware as a service." Newsworthy
- The glasses that told a dad to stop yelling - Viture's Vonder specs, announced Tuesday, ditch the display and the cameras to build a "vision board" of your memories and goals. CEO David Jiang says his pair flagged how he spoke to his son at tennis tournaments. Viture
- The robot clocked in without being carried - XPENG says its Iron humanoid walked off the production line on its own after assembly, earning a staff badge from the CEO. First jobs: store greeter and tour guide, with mass production still a year-end goal. engadget.com
- $103 billion in contracts, two customers - British neocloud Nscale heads to the NYSE at a reported $35 billion valuation, with 85% of its backlog coming from Microsoft and Anthropic - and Anthropic can walk if Nscale misses "stringent" milestones. techcrunch.com
- Both sides are right, says Dalio - Ray Dalio told the World Economic Forum AI will be "miraculous" for productivity and also inflate a bubble with "devastating effects." He points to the Nasdaq's 86% run in 1999 and its 77% collapse by 2002. fortune.com
Today's Takeaway:
OpenAI's quality-control layer is the humans reading your actual chat logs, and some of them are quietly delegating that judgment back to a chatbot (Bbc). That's the same brittleness MIT Technology Review found under this summer's biggest claims: the impressive part evaporated once people with domain expertise checked the work (technologyreview.com).
Verification is the scarce input now, not capability. Which is why Amazon's move against Muse reads less like a privacy defense than a trust problem it can't inspect - an agent it can't identify is an agent it can't verify (Ai Meta). If you're deploying AI, your bottleneck isn't the model's output. It's whether anyone on your team is genuinely reading it.
🔍 Myth Buster
The myth: "AI oversight means a human is checking the machine's work - so AI-generated content in the wild has been vetted by a person."
The reality: OpenAI's own contractors - the people paid specifically to rate ChatGPT conversations and catch bad outputs - are getting fired for quietly using AI to write those reviews, with one contractor calling it the fastest way to get kicked off the job (BBC). The 'human in the loop' can be the exact same chatbot loop you're trying to check, which is why MIT Technology Review found this summer's biggest AI claims fell apart the moment someone with real domain expertise actually looked closely.
The nuance: OpenAI catching and firing these contractors is itself evidence the controls aren't totally broken - the concerning part is that the layer built specifically to catch machine shortcuts is the one where people are taking them.
The Bottom Line
The Pattern: Since the spring, AI's credibility problems have been blamed on models. Today they land on staffing and contracts - fired contractors, a blocked agent, a $103 billion order book resting on two logos. Trust is becoming an org-chart line item.
The Other Read: Firing a few contractors for cheating is ancient news in outsourced labor - this happens in content moderation and data entry every year, and OpenAI catching them suggests the controls work. We lean the other way because the cheating happens inside the one layer built to catch machine output.
Your Move: Fifteen minutes today: take one AI-generated deliverable your team shipped last week and have a person who knows the subject check it line by line. Log how many errors they find. That's your real review budget.
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
✨ How AI-Fluent Are You? - Test your AI fluency with our interactive quiz. See how you stack up and discover what to learn next. → Take the Quiz
💬 Community | 📞 Book a Consultation | 🌐 Website
Fluently yours, The My AI Fluency Team