ChatGPT saw something no human could

Happy Sunday,

I found something on X this week that might be the craziest AI use case I've ever seen... someone dropped their ring on the ground, took a photo, and ChatGPT found it before a human could. I'm giving you the photo below so you can try to spot it yourself. I'm also going deep on a Stanford study that just proved AI agrees with you 49% more than a human would (and the one prompt that fixes it), my top 3 AI tools for every type of task, five hidden Claude connectors, a couple quick AI updates, and five Claude cheat codes I've been using daily.

Let's get into it.

🤯 THIS IS WILD

The Craziest AI Use Case I've Ever Seen

I was scrolling X a few days ago and came across this post that actually made me put my phone down for a second. Someone dropped their ring on the ground. Couldn't find it anywhere. So they did what any reasonable person in 2026 would do... they took a photo of the floor and sent it to ChatGPT with the message "I dropped my ring, can you help me find it?"

Here's the photo. I want you to try and spot the ring before I show you where it is. Take a real second and actually look...

Did you find it? I stared at this for an embarrassing amount of time. I even zoomed in. Nothing.

ChatGPT found it in about three seconds. It analyzed the image and then generated a new version with a red circle drawn around exactly where the ring was sitting. Here's the result:

Here's what's actually happening under the hood. ChatGPT's vision model doesn't "see" images the way we do. It breaks the image into thousands of small patches and analyzes color, texture, and shape patterns in each one. It's comparing what it knows about what a ring looks like against every tiny section simultaneously. So even when the ring blends into the background and our brains skip over it, the model catches the subtle difference in texture and reflection. Pattern recognition at a scale that's physically impossible for human eyes.

The fact that it then generates a new image with a circle drawn on it is the cherry on top. That's vision capabilities chained with image generation in one shot. Next time you lose something small... an earring on a carpet, a screw on a workbench... just take a photo and try it.

🧠 THIS CHANGES EVERYTHING

AI Agrees With You 49% More Than a Human

Stanford just published a study in Science that tested 11 major AI models across over 11,000 real interpersonal scenarios. What they found is that AI models affirm what you say 49% more often than a real human would. And not just harmless stuff... they found AI agreed with people who were being deceptive, making unethical decisions, and doing things that were clearly wrong. The participants who received sycophantic AI advice became more convinced they were right, less likely to apologize when they messed up, and rated the sycophantic AI as more trustworthy. So the more AI tells you what you want to hear, the more you trust it. Even when it's wrong.

I've noticed this myself. I'll ask about a business decision I'm already leaning toward and it basically just validates me every time. It took me a while to realize... I'm just talking to a mirror that sounds smarter than me.

The fix. This comes from Andrej Karpathy (co-founded OpenAI). He built something called the LLM Council. Instead of asking one AI model one question and getting one potentially sycophantic answer, you force the AI to answer from five completely different perspectives that naturally disagree with each other. Then they review each other's answers anonymously, and a final chairman synthesizes everything into one clear verdict. Think of it like asking five friends with totally different life experiences instead of one.

The five advisors:

1. The Contrarian – actively looks for what will fail in your plan

2. The First Principles Thinker – asks "what are you actually trying to solve here?"

3. The Expansionist – finds the upside everyone else is missing

4. The Outsider – zero context about your field, catches the obvious stuff experts miss

5. The Executor – only cares about what you're doing Monday morning

How to set it up. Open up ChatGPT, Codex, or Claude, start a new chat, paste in this file, and say "Set this up as a skill." That's it. From now on, every time you need a real decision pressure-tested, you just say "council this" and it runs the whole thing automatically.

I've been using this for every major decision for the past two weeks. Business investments, content strategy, even personal stuff. The answers are wildly more useful. Save this for the decisions that actually matter... hiring, big investments, pivoting your business, quitting your job... the ones where getting it wrong actually costs you something.

Get the full LLM Council prompt →

🧠 SAVE THIS ONE

Top 3 AI Tools for Each Task

A lot of you asked for something simpler than the full tool breakdown I gave a couple weeks ago. Just tell me the top 3 for each category. So here it is.

For writing, I start in Claude because it sounds the most like a real human. For learning, nothing beats NotebookLM right now... you upload a textbook, a podcast, whatever, and it turns it into something you can actually interact with. For business owners specifically... Claude handles your automations and workflows, Cytd AI is incredible for getting your business discovered by the right people online (it runs a free AI visibility check that shows you exactly where you're being found and where you're invisible), and Manychat converts those leads on autopilot through Instagram DMs. Those three together are a full AI growth stack.

And for building apps, I'd look towards Claude Code or Codex now. They're both good at different things so I'd suggest using both... but if you're completely non-technical, Lovable is definitely the move. It's what my website is built on and honestly one of my favorite AI tools that's ever been released.

Get my full AI tools breakdown →

🔗 HIDDEN FEATURES

5 Hidden Claude Connectors

Connectors are one of the most powerful things in AI right now. They let Claude control other apps directly inside the software. Here are my five Claude connectors that actually make a difference. Claude can pull on these whenever you need them as long as they're connected.

1. Canva
Tell Claude what to design and it creates it directly in your Canva account using your brand colors and templates. I've used this for pitch decks and the designs are honestly better than what I would've made myself.

2. Meta Ads
Claude pulls data directly from your ad account, finds trends you missed, and can suggest or launch new campaigns. This one isn't an official Claude connector... you connect it through MCP. If you're curious about setting it up, check out this guide.

3. Booking.com
Tell Claude where you want to travel and when. It searches in real time, compares hotels based on your preferences, and books the entire stay. I used this for a trip last month and it found a better deal than what I was about to book manually.

4. Higgsfield
Generate images, videos, and UGC-style ads from all the top AI models without leaving Claude. This one is also connected through MCP, not the official connector menu. If you want to set it up, check out this guide.

5. Gmail
Claude reads your inbox, drafts replies in your voice, and clears out subscriptions. Still my most-used connector. If you missed the newsletter where I set this up, go back and read it.

All of these are set up through Customize → Connectors inside Claude. Most take less than two minutes.

Get the full connectors setup guide →

📰 QUICK UPDATES

Two Quick AI Updates This Week

1. Apple is finally making Siri useful.

Apple is announcing a completely rebuilt Siri at WWDC this week, powered by Google's Gemini model. It can understand context, handle multi-step tasks, and access your emails, photos, and files. But here's what matters most... Apple is introducing an Extensions system that lets you choose between ChatGPT, Gemini, and Claude as the AI model powering features on your iPhone. All of these tools we talk about every week are about to be built directly into every iPhone on the planet. If you're already here learning this stuff, you're going to be way ahead of that wave.

2. ChatGPT's memory just got a massive upgrade.

OpenAI rolled out "Dreaming V3" this week. Before this, ChatGPT would save little facts you told it. Now it reads across ALL your past conversations in the background and automatically builds a full picture of who you are, what you're working on, and what you care about. It even updates memories over time... if you told it you were planning a trip in July, once the trip ends it automatically updates so it stops giving you travel tips. Every conversation is about to get way more personalized without you repeating yourself. Rolling out to Plus and Pro users first.

🔒 CHEAT CODES

5 Claude Cheat Codes That Actually Work

First I gave you 10 secret codes for ChatGPT. A bunch of you asked if Claude had similar ones. It does. These aren't official Claude features, just commands that change how Claude responds to you. Here are the five I use the most.

/compact – Works inside Claude Code. Compresses your entire conversation history so you don't hit token limits and get cut off mid-task.

SWAT – Add this to the end of any prompt and Claude runs a full strengths, weaknesses, opportunities, and threats analysis on whatever you asked about.

/premortem – Claude pretends your plan already failed six months from now and works backward to tell you exactly why. I've killed multiple ideas after running this.

STEELMAN – Claude argues the absolute strongest version of the opposing side. Incredible for debates, negotiations, or understanding the best counterargument to your position.

REDTEAM – Claude attacks your plan from every angle a critic, competitor, or skeptic would. If your idea survives a red team, you know it's worth pursuing.

Get 25 more Claude cheat codes →

Go try that LLM Council prompt on a real decision you're sitting on right now. I promise the answer you get back is going to be completely different from what you're used to. The moment you realize AI has been agreeing with you this whole time is a weird feeling... but the fix is worth it.

See you Wednesday.

Maverick

P.S. If you're a business owner looking to actually implement AI into your workflows, my team and I work with a small number of people and businesses to build custom AI systems, automations, and tools. If that sounds like you, apply here.

P.P.S. If you're new here or you ever skip one, I've got a newsletter archive of all my past emails you can catch up on anytime.