Claude can now think 100 different ways

Happy Wednesday,

Today I've got four image codes that turn an ordinary photo into something completely different, the FBI arrest over a hidden recorder inside OpenAI, and a skill that gets Claude to compare different approaches before giving you its best answer.

Let's get into it.

💡 SAVE THIS

4 Secret Image Codes for ChatGPT

A few weeks ago I shared some image codes. Here are four more to try on your own photos.

Upload a photo into ChatGPT and type one of the codes below. One has you walking through the frame like a ghost, one gives you the paparazzi treatment, and another looks like someone photographed you from the floor with a very strange lens.

Use the same original photo for each one if you want to compare the results. Starting a fresh chat for each style also keeps it from mixing the last effect into the next one.

1. /ghost-walk This gives you that blurred, almost see-through movement, like the camera caught several moments of you walking in one shot. Try a full-body photo with some space around you so the effect has somewhere to go.	2. /paparazzi-editorial Think walking out of a restaurant and getting photographed before you're ready. A little chaotic, a little dramatic, and considerably cheaper than becoming famous.
3. /worms-eye-wide This is the one with the oversized shoes, stretched perspective and buildings bending around the edges. A standing photo gives it more to work with than a tightly cropped selfie.	4. /fire-fade You get the dramatic fire effect around the image while still looking like yourself. If it goes a little too far, ask for fewer flames and more glowing embers.

You can keep talking to it after any of these. Say “keep my original outfit,” “make the effect more subtle,” or “use the same lighting as my original photo.” You don't need to get the perfect result on the first try.

I've put 200 more image and video shortcuts in the guide below, including product photos, ads, illustrations and plenty of other things to try.

Get my 200 ChatGPT image and video codes →

🚨 THIS IS WILD

The FBI Just Arrested an OpenAI Employee

Someone allegedly hid a recorder behind a TV at OpenAI.

The FBI arrested an OpenAI engineer on October 9 after a recording device was found inside an executive conference room at its San Francisco offices. Prosecutors announced the case yesterday, and the details sound like something out of a spy movie with a very questionable props budget.

According to the complaint, the recorder was plugged into the back of a television and concealed in toilet paper. OpenAI found it on September 30. Prosecutors say the employee later came back and looked behind the TV, and security staff found a second similar recorder in a bag containing his wallet and ID. These are allegations, and he hasn't been convicted.

Then there's the part that makes this a very 2026 story. Prosecutors say a chatbot account associated with the employee's spouse had been used to ask about building and hiding a recorder. And OpenAI says its own AI helped examine the device and trace where it came from, helping identify the employee.

So AI may have helped someone figure out how to hide a recorder, then helped the company figure out who put it there. OpenAI says it fired the employee and has no indication that customer data, source code or its trained models were improperly accessed or taken.

People on Reddit are arguing about whether this shows how useful AI has become or whether OpenAI managed to turn an FBI arrest into a product demonstration. I think both can be true. If AI helped them investigate, that's useful. It doesn't mean ChatGPT solved the case by itself. The company hasn't explained enough to know how much of the work its models actually did.

What gets me is the contrast. One of the companies building the most advanced technology in the world is dealing with a recorder allegedly wrapped in toilet paper behind a television. You can spend billions making your AI smarter and still have a problem that comes down to who can walk into a room.

Read the case details →

🧠 TRY THIS

Get Better Answers Out of Claude

If Claude keeps giving you bad answers, this one is worth trying.

I came across a skill on GitHub called the Arena skill, and I decided to test it. The results are pretty wild.

So essentially, you give it one task, and it spins up a group of 100 Claude agents to work on that exact task, each taking a different approach.

They compare answers, point out what's weak and improve their work through a series of rounds. The strongest answer moves forward until you're left with a winner. So instead of asking the same chat to “make it better” over and over, you're getting different ideas tested against each other.

That much Claude working on one problem will burn through your usage, even though the skill itself is free. I'd start with 15 agents and save the bigger run for something you're really stuck on.

But if you've got a big problem you've already spent hours on, sending in all 100 can be worth it. Especially if you're already on Claude Max 5x ($100/month) or Max 20x ($200/month), which give you more usage than Pro. They're still capped, so the idea is to spend that extra capacity on something that matters.

Think about a proposal that sounds too generic, a headline you can't get right, or a plan where you'd like to compare a few different directions. Those are the kinds of tasks where the extra effort could save you another afternoon of going back and forth.

Here's how to set it up.

Step 1: Open Claude Code in the Claude desktop app. You'll need a paid plan that includes Claude Code. You can use it for writing and planning too, so don't let the name put you off.

Step 2: Paste this link and ask Claude to install the skill:

Once it's installed, you can use /arena whenever you've got something you want it to work through.

Step 3: Give it one clear job and set it to 15 agents. Here's a starting point:

For a bigger problem, change --agents 15 to --agents 100 in that same prompt.

Give it the context you'd give a person helping you. Who is this for? What needs to stay? What didn't you like about the first version? The clearer you are about those things, the more useful the different approaches will be.

You still get the final say. Claude is judging the answers too, so check the winner before you use it. But if you keep getting slight variations of the same answer, this gives it a reason to try something different.

I've put the full setup and more examples in the guide below.

Get my full Claude Arena setup guide →

Try one of the image codes on a photo you already like, and reply with what you get. And if Claude has been circling the same answer all week, give the smaller Arena run a try before sending in all 100.

See you Sunday.

Maverick

P.S. Want your business to actually show up inside AI? The brands that rank are seeing up to 3x the conversions. Get your free AI visibility score and see exactly where you stand inside AI.

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.