|
Happy Sunday everyone,
This week was insane. Anthropic just announced a new Claude model so powerful they refuse to release it to the public. A crazy prompt chain that gets ChatGPT to reveal stuff about you that you've never said out loud. A new AI UGC tool that's letting people make viral ads. And the next ChatGPT image model just leaked and it looks wild.
Let's get into it.
|
IN THIS EMAIL
» The AI model that's too powerful to release
» The ChatGPT reveal prompt chain
» The next ChatGPT image model just leaked
» Tool of the week: Poppy AI
» ChatGPT lies to you 27% of the time (and the truth prompt that fixes it)
» Play this AI game right now, thank me later
|
📰 THIS WEEK'S RABBIT HOLE
The AI Model That's Too Powerful to Release
Anthropic (the company behind Claude) just built a brand new AI model called Claude Mythos. They're saying it's the most powerful AI they've ever made by a huge margin. Then they told the public we can't have it.
The reason? It's so good at finding weaknesses in software that they're worried about what would happen if it fell into the wrong hands. So instead of releasing it to everyone, they're giving it to a handful of tech companies and security experts through something called Project Glasswing. The idea is to use it to find and fix the biggest security holes on the internet before anyone else builds something like this.
This is the first time I can remember one of the major AI labs basically saying "our model is too dangerous to release" and actually meaning it. We're in a new era.
Read Anthropic's full announcement →
🎥 VIDEO YOU MIGHT HAVE MISSED
The ChatGPT Reveal Prompt Chain
I tried something this week and the results were actually terrifying. What happens when you ask ChatGPT to reveal everything it knows about you based on every single conversation you've ever had with it?
You need to run these 4 prompts one after the other in the same chat. Here they are:
Prompt 1:
|
Tell me everything you know about me from our past conversations. Include my personality type, my insecurities, my goals, my fears, and what you think I actually want in life.
|
Prompt 2:
|
Tell me what my biggest blind spots are and what patterns you've noticed that I probably don't even see in myself.
|
This one literally flagged things about me that I've never said out loud.
Prompt 3:
|
What is the single biggest thing holding me back right now? And what would my life look like in 5 years if I never fixed it?
|
ChatGPT will then give you a full 5-year life prediction. Honestly, mine was something I wasn't ready to hear.
Prompt 4:
|
Now flip it. If I fix that one thing, what does my life look like in 5 years instead? Be specific.
|
ChatGPT is going to paint you two completely different pictures and show you exactly what happens if you do or don't take action. It's almost scary how accurate it gets.
Fair warning: you need to have had a decent amount of conversations with ChatGPT for this to work properly. If you're new to it, the results won't be as accurate. But if you've been using it for a while, try this one today. Takes 5 minutes and you might learn something about yourself.
Watch the full video on Instagram →
👀 LEAK OF THE WEEK
ChatGPT's New Image Gen Model Just Leaked
Both of these images aren't real. They were generated with the new model.
OpenAI's next image generation model (the thing that replaces the current image creator inside ChatGPT) just got leaked on a public testing site earlier this week. People spotted three mystery models with weird code names, ran a bunch of prompts through them, and the outputs are insane.
The biggest jump is in text. If you've ever tried to get ChatGPT to make an image with real readable text on it, you know it's usually a mess. This new model basically solved that. People have been generating fake websites, fake UI screenshots, product labels, billboards, signs, all with perfectly clean text. The realism is a huge step up too.
OpenAI hasn't officially confirmed anything, but the model was pulled from the testing site shortly after it got noticed, which usually means a release is close. Expect this to drop sometime in the next few weeks.
🛠 MAVERICK'S TOOL OF THE WEEK
Poppy AI
This tool is literally game changing. It's called Poppy and you've probably heard me talk about it on my channel before. I know I mention a lot of tools, but Poppy is one I ACTUALLY use every single day throughout my entire content workflow. My whole content team is on it too. I literally forced them all to onboard because it's that good.
It gives you a whiteboard canvas where you can drag in any type of content (YouTube, TikTok, Instagram, Meta ads, voice recordings, pretty much anything) and connect it to an agent running on Claude or GPT. This lets you feed the agent full context from whatever you drop in, so you don't have to waste time pulling scripts or manually copying stuff off other videos. Then the magic happens.
Here's the part that makes Poppy different from every other tool out there: it can actually understand the videos you give it. So if you drop in a YouTube video, it doesn't just pull the script like every other tool on the market. It can literally SEE the video. It picks up on the visuals, the style, the pacing, the b-roll, everything that's actually happening on screen. That's a massive deal if you care about writing scripts or breaking down why certain videos actually work. They also have all the major AI models and image generation built right in.
Quick heads up: it's not free and a little on the expensive side. But as someone who writes scripts off a ton of other videos, I'd be nowhere without it. If you create content, this is the one.
Try Poppy AI with my special link (discount) →
I also filmed two tutorials on how I actually use it:
» Poppy AI tutorial #1
» Poppy AI tutorial #2
⚠️ IMPORTANT TO KNOW
ChatGPT Lies to You 27% of the Time
A Deakin University study found that nearly 3 out of every 10 things GPT-5 generates are straight up fabricated. That means roughly 1 out of every 4 things ChatGPT tells you might be wrong, and it says it all with the same confidence as the stuff it gets right. You'd never know unless you go check.
There's a way to force it to stop doing this. It's called the truth prompt. You paste it into your custom instructions inside ChatGPT and it sets a bunch of ground rules: no guessing, no making things up, every claim has to be backed by a real source. To install it, grab the prompt here, then open ChatGPT, go into personalization settings, click custom instructions, paste it in, and hit save.
Here's a video tutorial as well if you want to follow along:
Watch the truth prompt video →
Source: Journal of Medical Internet Research hallucination study
🎮 JUST FOR FUN
Play This Right Now, Thank Me in 30 Minutes
Google has a free browser game called Quick, Draw! and it's one of the most fun AI things I've come across. It gives you 20 seconds to draw something (like a banana, a squirrel, a pair of sunglasses), and an AI has to guess what you're drawing in real time as you're doodling it. The crazy part is watching the AI narrate out loud what it thinks you're drawing. It'll say "I see a cat... no wait, a horse... oh a giraffe!" as you're still sketching. It's hilarious.
It's also a really simple way to actually understand how AI learns. Every drawing you do teaches it to recognize shapes better, and Google has been using millions of these doodles to train models for years. It's one of the original AI experiments that's still just as fun today as it was when it came out.
Free, no signup, open it in your browser and go. I promise if you play it right now you will thank me in 30 minutes.
Play Quick, Draw! →
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. Missed a past email? My full newsletter archive is always there for you.
|
