AI DUNN Right Weekly - Issue #10
- Dec 1, 2025
- 6 min read
Practical AI insights for business growth
Hey AI Innovators!
Welcome back to AI DUNN Right Weekly! This week brings a leaner, meaner newsletter format based on your feedback. We're cutting the fluff and keeping what matters: real AI news with my honest take, tools you can actually use, and prompts that work.
This week OpenAI made ChatGPT smarter AND more human, Google's Gemini 3 Pro is crushing benchmarks while Anthropic launched its fastest premium model yet, and Perplexity just turned shopping into a conversation. Plus, the $50 billion AI infrastructure race that's reshaping where and how AI actually runs.
Here's what matters for your business.
Read time: 4 minutes
This Week's Game Changer
ChatGPT Gets Personality Control - And It Actually Matters
What happened: OpenAI rolled out GPT-5.1 with two modes (Instant and Thinking) and something that sounds minor but ain't: full personality customization. You can now pick preset vibes like Professional, Friendly, or Nerdy, then dial in exactly how warm, concise, or emoji-heavy you want responses. These settings stick across all your chats. (Source: OpenAI Release Notes, BinaryVerse AI)
Why it matters: For the first time, ChatGPT feels less like a corporate robot and more like a colleague who actually knows you. If you're using AI for customer service, this means your responses can finally match your brand voice without constant prompt gymnastics. If you're a solopreneur juggling code help and business strategy, you ain't getting the same stiff tone for both anymore.
Here's what I think: I've wasted hours fighting with ChatGPT's tone. It sounds like a robot when I'm trying to brainstorm creative ideas, then gets way too casual when I need professional client emails. Having to re-prompt "be more professional" or "be more creative" every single time is exhausting. Now you set it once and it sticks. That's the difference between a tool you fight with and one that actually helps you work.
Business impact:
• Customer service teams can embed brand voice at scale
• Freelancers can switch between client tones without retraining
• Marketing teams get consistency across dozens of campaign drafts
• Developers get faster responses when they don't need deep reasoning (GPT-5.1 thinks harder only when the prompt deserves it)
What to do: If you're on ChatGPT Pro or Plus, go into settings and experiment with tone presets. Test how "Professional + Concise" hits different than "Friendly + Detailed" for your actual work. You'll find your sweet spot fast.
AI Tool Spotlight
Perplexity's Instant Buy - AI Shopping That Doesn't Suck
What it does: Perplexity partnered with PayPal to let you find products, compare options, and check out, all inside the chat. No tab-hopping, no getting lost in sixteen browser windows. Ask for "gaming laptops under $1,500" and you get recommendations with instant buy buttons. First purchase through PayPal gets you 50% cashback (up to $50). (Source: AllAboutAI, Perplexity)
Why you care: This ain't just another chatbot gimmick. This is AI actually closing the loop on commerce. You go from "I need this" to "I bought this" without leaving the conversation. Newegg's already integrated, so tech purchases are live right now.
Here's what I think: We've all done this. AI recommends the perfect product, you open five tabs to compare prices, read reviews, check shipping, forget what you were even looking for, and end up buying nothing. Perplexity just killed that entire waste of time. You ask, you compare, you buy. Done.
The 50% cashback is obviously marketing, but the smart part is PayPal keeping retailers as the merchant. Brands still own the customer relationship, so they'll actually want to integrate this instead of fighting it.
Who should use it:
• E-commerce teams watching where conversational commerce goes
• Tech buyers who value speed over exhaustive research
• Anyone tired of the Amazon search-scroll-compare death spiral
Real-World Win
Claude Opus 4.5 - The Productivity Benchmark That Actually Ships
What happened: Anthropic launched Claude Opus 4.5 on November 25 with 80.9% accuracy on SWE-bench (the gold standard for AI coding tests). It's faster than Claude Sonnet 4 and costs 67% less at $5 per million tokens. It includes "thinking mode" for complex multi-step tasks and agentic plugins for workflow automation. (Source: Recent AI Model Releases, HumAI Blog)
The proof:
• Early adopters report it handles day-long refactoring projects without losing context
• Design teams praise its aesthetic judgment on spacing, typography, and white space (stuff previous models botched)
• One developer said it "feels like it actually understands what I'm building, not just executing commands"
Here's what I think: Most AI launches give you one thing - faster OR cheaper OR smarter. Pick one. Anthropic actually delivered all three with Opus 4.5. That's rare and that's why businesses will actually pay for it.
The "thinking mode" is the real winner here. You're not burning money on deep computation when you ask "what's the weather?" but when you need it to work through a complex refactoring project, it's there. That's how pricing should work - you pay for what you use, not flat-rate expensive overkill.
Business impact:
• Development teams can ship faster
• Design agencies can prototype with AI that doesn't create visual disasters
• Consultants can draft complex proposals without babysitting every output
• The cost reduction means more companies can afford to use premium models in production, not just for demos
Prompt of the Week
The "Structured Thinking" Framework for Complex Decisions
Use this when you need AI to work through product strategy, operational decisions, or multi-variable problems:
Why it works: This forces ChatGPT (or any LLM) to show its work instead of jumping to conclusions. You catch bad assumptions early, and the breakdown makes it easier to spot where the AI might be off track.
Example use: "I need help analyzing whether to hire a full-time designer or use freelancers for our Q1 campaign." The prompt walks through budget constraints, stakeholder needs (marketing team, clients, internal brand), trade-offs (quality vs. flexibility), and downstream effects (team morale, brand consistency).
Here's what I think: Most people use AI like a magic 8-ball - shake it and hope for a good answer. This prompt forces it to actually think through the problem step by step, like a real consultant would. The "don't skip ahead" part matters because AI loves jumping to conclusions without showing its work.
Quick Hits
Three More Things You Should Know:
1. The $50 Billion Infrastructure Race Amazon (AWS) is investing $50 billion to build AI data centers for federal agencies, adding 1.3 gigawatts of compute capacity. Microsoft formalized a $135B stake in OpenAI with compute tied to Azure through 2032. Meta's guiding $70-72B in 2025 capex for AI infrastructure. (Source: State of AI Newsletter, AI News November)
Here's what I think: These companies are betting everything on owning the infrastructure where AI actually runs. AWS going after federal contracts first is brilliant ...add government security standards, then every enterprise follows. The dollar amounts are insane, but what matters is these three companies are fighting to control the entire compute layer. Whoever wins this gets to decide how expensive AI is for everyone else.
2. Gemini 3 Pro Tops Leaderboards Google's Gemini 3 Pro launched November 18 and immediately hit #1 on LMSYS Arena. It solved 5 of 6 IMO 2025 math problems in "Deep Think" mode and was trained exclusively on TPUs. Integrated into Search, Maps, and Vertex AI. (Source: AI News November, Recent AI Model Releases)
Here's what I think: Google's been getting embarrassed by OpenAI and Anthropic for months. Gemini 3 Pro is their "we're still here" moment. Training it only on their own TPUs instead of Nvidia chips is a power move (they're proving they don't need anyone else's hardware). If you're already using Google Workspace, this matters because AI is now baked into Search and Maps where you actually work.
3. Deloitte's AI Citation Scandal A Canadian province asked Deloitte to review a $1.6M health report that contained fabricated citations and errors from AI-generated content. Similar incident in Australia where Deloitte partially refunded a client for made-up academic references. (Source: The Latest AI News)
Here's what I think: If you're using AI for client work without checking every single output, you're gambling with your reputation. Deloitte's saying "the conclusions are right, we just made up all the evidence" (that's insane). One fake citation can get you sued and destroy years of trust. This is why "AI without human verification" is a disaster waiting to happen for consultants and agencies.
That's it for Issue #10!
We've streamlined the format based on your feedback - fewer sections, more substance, and my actual opinions on what matters. Let me know what you think about the new structure.
Stay innovative,
Jackie @ AI DUNN Right
P.S. We're now 340+ subscribers strong. Thanks for growing this with me.
Got AI questions? Hit reply. I read every email.
Want to work with Kyle Balmer's AI training? Check out his workshops - he's doing practical implementation training for teams.
Share this newsletter: [Forward to a colleague who needs better AI insights]
Comments