🍭 Apple: AI Sucks 🤷‍♂️

Good morning. It’s Wednesday and Apple’s claiming reasoning models aren’t reasoning.

Meanwhile, someone in our circle booked 3 deals using one single build.

Let’s dig in.

Reading Online?
→ Subscribe here ←

🍭 What’s Cookin’:

• AI doesn’t “think”

• Midjourney catching strays from Hollywood

• 5 AI tools to help you build, price, and manage your project

• And Everything Else….

Apple
🧠 Apple: "AI Doesn’t Think.” kinda…

The Bite: In a paper titled “The Illusion of Thinking”, Apple claims that top reasoning models (Claude 3.7, DeepSeek R1, o3 Mini) don’t actually reason.

They claim the models just memorize patterns until things get too hard.

The internet lost it.

But the whole thing might be more about context window limits than some fundamental brain bug.


Snacks: 

🍭 Apple tested “large reasoning models” (LRMs) on logic puzzles like Tower of Hanoi

🍭 Accuracy drops to zero past a complexity threshold — even with compute left

🍭 Models used fewer tokens on harder problems

🍭 They failed even when fed the correct algorithm

🍭 Critics say models weren’t “thinking less” — they just hit token/context window limits

🍭 Apple blocked tools like coding, which solve the puzzles instantly

Why it bites: Apple wrote a paper saying the duck might not be a duck.

But it walks, quacks, and automates your operations just fine.

The real twist?

This was less “AI can’t reason” and more “AI hits its word count.”

The models didn’t break — they literally knew they’d break and told us to just use the algorithm.

That’s not failure.
It’s a setup.

Ironically, they dropped this days before Apple’s WWDC25 (covered yesterday), where they announced the most AI integrations into Apple’s OS…

Teaser: We’ll tell you how OpenAI responded to this later this week 😉
Time for things to get interesting.

🍭 Steal this Prompt
👑 Turn any Prompt Into This 👇

🎯 Built for: Print posters, streetwear mockups, album covers, or just flexing on Twitter
🎨 Output: Colorful, minimalist vector-style illustrations
⚠️ Midjourney only!

📌 Tip: Use this prompt with strong nouns (skeletons, UFOs, flowers, swings) for the biggest punch

1. Get the prompt here

2. Paste into Midjourney

3. Swap in your #prompt with any prompt you want to create with

🍭 Try the prompt on Snack Prompt 🍭

DeFAI
🎬 Disney/Universal Sue Midjourney: “AI Piracy Is Still Piracy”

The Bite: Disney and Universal just sued Midjourney for copyright infringement.

The first major Hollywood strike against an AI image generator.

The 110-page suit accuses Midjourney of mass-plagiarizing characters like Darth Vader, Elsa, and Homer Simpson by training on their images without permission.


Snacks: 

🍭 Midjourney “helped itself to countless” copyrighted works, per the lawsuit

🍭 Characters cited include Shrek, the Minions, and Disney Princesses

🍭 Studios claim AI theft “threatens U.S. copyright law’s foundation”

🍭 Disney previously sent a cease & desist — Midjourney ghosted them

🍭 Midjourney made ~$300M last year selling image gen subscriptions

🍭 Studios now want to block its upcoming video gen feature, too

Why it bites: Remember when Epic Games got heat for using Darth Vader’s AI voice in Fortnite instead of paying the OG actor?

This is the next logical boss fight.

Except this time it’s not actors suing studio.
It’s studios suing the toolmakers directly.

Hollywood’s finally stepping in… not to protect creators, but to protect their IP vaults.

Which, to be fair, is worth billions.

Together with Recraft
🖼️ Consistent Visual Styles with Recraft

For anyone tired of endlessly scrolling through stock photos and editing them to fit their brand — meet Recraft. It’s an AI-powered image generation and editing tool. The recent update, Advanced Style Creation and Control, makes it much easier to build and maintain consistent visual styles across your work.

Why Recraft?

Trusted by 4 M+ users and teams at Netflix, Ogilvy, and Hubspot, Recraft empowers creators to:

🎨 Explore endlessly with the Infinite Style Library

⚡ Work faster with AI-native tools

🧩 Stay on-brand by saving and sharing styles across teams

From a single prompt and style reference, generate logos, illustrations, icons, mockups, and more, fully aligned with your brand.

Special offer for SnackPrompt’s subscribers: use promo code SNACK11 for a $11 discount on any paid plan!

Big Biz
🔁 One Build. Many Wins.

The Bite: The first time you build an automation, it feels like a project.

By the third time, it’s pure leverage.

One builder in our orbit reused the same onboarding flow.

Logo swap, minor tweaks, one fresh email and landed three wins with zero rebuilds.

Snacks:

🍭 Onboarding never goes out of style

→ Every business wants better client onboarding.

🍭 Repetition = Speed

→ The more times you deploy the same system, the faster you get.

🍭 Clients don’t need novelty

→ They want something that works. Period.

Why it Bites: Too many solo builders try to make everything custom.

But the real pros?

They template once and stack wins on top.

You don’t need more builds.

You need one that multiplies.

Everything Else
🧠 You Need to Know

OpenAI’s open model delayed until “later this summer”

→ Sam Altman says the company’s first open-weight model is taking longer than expected — but promises it’ll be “very, very worth the wait.”

O3-Pro: OpenAI’s new model beats Claude and Gemini

→ The o3-pro upgrade outperforms rivals on math, science, and reasoning — but it’s slower, pricier, and still can’t generate images or use Canvas.

Meta unveils V-JEPA 2, a “world model” for physical reasoning

→ Trained on 1M+ hours of video, V-JEPA 2 lets AI agents reason about real-world actions — think robots that actually know how to plate eggs.

Zuck hires Scale AI’s CEO in $14B power move

→ Meta’s massive investment brings Alexandr Wang aboard to reboot its GenAI efforts after LLaMA 4 flopped with devs — and rivals pulled ahead.

Tool Box 
🧰 Trending AI Tools

🧮 Atlas – Pricing pages that print money

Get plug-and-play billing setups with zero dev grind. Launch today, monetize tomorrow, vibe instantly.

🧠 Merlio – One command center to rule all your AIs

Tap GPT, Claude, Gemini & more from a single power tool for content, images, and idea gen. AI fusion core unlocked.

✂️ undash – Kill em-dashes, save humanity

This Chrome plugin auto-strips long em dashes from AI text on paste. No setup. No drama. Just clean copy.

🕹 RetroTerminal – Gemini in a neon-drenched terminal

Chat with LLMs like it’s 1984. Old-school UI, modern power. Local memory, full theme support, and total API control.

🕸 Spydr – Context memory that follows you

GitHub for LLM memory. One context engine, every app. Finally, your AI doesn’t forget who you are mid-convo.

Referral Program
Invite Friends & Earn Some Freebies 🔍