Hello and welcome to the Automated, your AI tour guide.

This weekend, AI news went wild. Meta dropped new models, Microsoft rolled out fresh updates, and OpenAI made some big moves, including delays and open-source plans.

It’s safe to say, the AI race is in full throttle!

Here's what we have for you today

• 🤯 Meta’s Saturday Surprise and OpenAI’s Latest Mo …

• 🤔 Real Proof That OpenAI’s Models Are Doing Mor …

• 🧱Around The AI Block

• 🛠️ Trending Tools

• 🤖ChatGPT Prompt Of The Day: Prompts for Brainstor …

🤯 Meta’s Saturday Surprise and OpenAI’s Latest Moves.

This weekend was a whirlwind of AI updates. Meta didn’t just drop its new flagship model, Llama 4—they dropped it on a Saturday.

But while we were still reeling from the new release, things got complicated fast—like, Meta finding themselves on the bad side of developers due to a blunder they may have intentionally made with one of their new AI models, Maverick.

Here’s what happened:

Maverick, one of the models in Meta’s new release, ranked second on LM Arena (an AI test where human raters choose their favorite model outputs)—and it was pretty impressive.

That is, until we found out that the version Meta showcased on LM Arena wasn’t the same as the one available to developers. 

Yep, you read that right.

Turns out the version on LM Arena was an “experimental” one—super chatty, full of emojis, and prone to long-winded answers.

But the version developers actually get? Not nearly as flashy…

Honestly, it’s like Meta gave the AI a makeover for the test, then sent it out into the world looking... basic.

Researchers on X are calling it out, highlighting how misleading that move was—especially since it makes it hard for developers to know how well the model will actually perform in real-world applications.

And they’ve got a point. When a model behaves one way in testing but another way in the wild, it’s like buying a car that runs like a dream in the showroom but stalls every time you take it for a spin. 

Not exactly confidence-building, right? Well, that’s what we got.

But that’s not all—OpenAI is making its own moves, too. 

After pulling the plug on the consumer launch of its o3 reasoning model back in February, OpenAI is now gearing up to launch both o3 and its successor, o4-mini, in just a couple of weeks.

And if that wasn’t enough, the much-anticipated GPT-5 has officially been delayed.

Sam Altman, OpenAI’s CEO, said it’s taking longer than expected to get everything working perfectly, especially with all the new features they’re packing in.

In case you forgot, GPT-5 is supposed to be the unified model that’ll take AI to the next level—think voice, deep research, Canvas, and search all in one.

Since GPT-5 is running behind schedule, OpenAI will fill the gap with o3 and o4-mini to keep the momentum going.

But here’s the kicker—OpenAI is going open-source again. After keeping things tight-lipped post-GPT-2, they’re set to release a new open language model with reasoning capabilities.

It’ll be interesting to see how this plays out, especially with competitors like DeepSeek pushing models out faster and making them open for anyone to tinker with.

So yeah, the AI launch game is in overdrive. Between Meta’s Saturday surprise and OpenAI’s never-ending updates, we’re left scrambling to keep up.

⏳ Final call: The Prompt Me bundle disappears at midnight!

Tonight is the final day of Prompt Me Week.

If you’ve been thinking about going Premium — this is your moment.

When you upgrade today, you’ll unlock:

• 📽️ 7 exclusive video prompt lessons (worth $14.99)

• 🧠 Daily AI prompt videos — like having an AI coach in your inbox

This bonus bundle won’t be available again. And enjoy a little trailer of our AI Workout of the Day feature - and like and subscribe!

So hurry and Join Premium now !

🔧 AI Troubleshooter: When ChatGPT "Forgets" Your Conversation.

Hey AI fans! Got another quick fix for you.

One of our readers wrote in, frustrated that ChatGPT kept “forgetting” parts of their conversation— especially during longer chats. super annoying, right?

Here’s how to fix it:

• ✅ Keep It Short: Break your prompts into smaller chunks and remind ChatGPT of important details as you go. This helps the model stay focused and on track.

• ✅ Turn on Memory: ChatGPT now has a memory feature! To enable it, head to Settings > Personalization > Memory, and toggle it on. This will allow ChatGPT to remember facts about you across sessions — like your preferences, goals, and even your favorite topics.

• ✅ Use Custom Instructions: You can also use Custom Instructions to tell ChatGPT how you want it to respond (like your preferred writing style or company name) and what it should keep in mind during a session. Just go to Settings > Custom Instructions and fill out the fields with your preferences.

• ✅ Start Fresh If Needed: Even with memory on, if a single conversation gets too long, it’s sometimes better to start a new chat to keep it running smoothly.

And Note: ChatGPT's memory and custom instructions features are available to most users—except in Europe and South Korea due to data protection regulations.

If you’ve got your own ChatGPT issue, fill out this quick Google Form, and we’ll send you a custom fix—just don’t forget: premium subscribers are prioritized'.

Now onto the next big piece of AI news…

 🤔 Real Proof That OpenAI’s Models Are Doing More Than Just Learning!

OpenAI's been swearing up and down that it’s playing fair with copyrighted content — but a new study just dropped the receipts, and let’s just say it’s not looking great.

Researchers from the University of Washington, Stanford, and the University of Copenhagen found a clever way to check if OpenAI’s models (like GPT-3.5 and GPT-4) were straight-up memorizing stuff they probably shouldn’t have seen.

And guess what? Your favorite AI might be guilty of a little literary sticky fingers.

Here’s how they did it:

The researchers used something called “high-surprisal” words — aka rare, somewhat weird words that don’t usually pop up in most texts. (Think “radar” in a cozy fiction sentence, where “radio” or “engine” would’ve made more sense.)

They removed these rare words from snippets of fiction books and New York Times articles, then asked the models (GPT-3.5 and GPT-4) to guess what was missing.

If the model guessed right? That’s a sign it didn’t just learn patterns — it actually memorized the content, holding on to exact phrases from its training data.

And yep, GPT-4 passed that test a little too well, especially with snippets from copyrighted fiction books and — to a lesser extent — articles from the New York Times.

So, basically, these models might be walking around with full paragraphs from books stored in their digital brains. Yikes.

One of the researchers even called this a wake-up call, saying that if we want AIs we can trust, we need to be able to poke, prod, and audit what they’ve been trained on. Makes sense, right?

Meanwhile, OpenAI’s playing both sides.

They’re making deals with content creators, offering opt-outs, and trying to look responsible.

But at the same time, they’re lobbying hard to make it 100% legal to train AI on copyrighted material without permission. Because apparently, "fair use" is their new weapon of choice.

At the end of the day, this study is another loud reminder: we need way more transparency about what these AI models are really trained on.

Otherwise, we’re just guessing — and frankly, the AIs seem to be way better at guessing than we are.

Full article is right here.

🧱Around The AI Block

• 👍 Top 11 AI form builders.

• 👨‍🏫 How to use AI to track your startup’s KPIs, enhance ops and impress investors.

• 🥳 All of the news from Microsoft’s 50th Anniversary Copilot event.

• 🤑 GitHub Copilot introduces new limits, charges for ‘premium’ AI models

• 😱 Microsoft employee disrupts 50th anniversary and calls AI boss ‘war profiteer’.

• 💸 Gemini 2.5 Pro becomes Google’s most expensive AI model yet.

• 🦾 Microsoft releases AI-generated Quake II demo, but admits ‘limitations’.

🛠️ Trending Tools

• Retrodiffusion Is a pixel-art style image generator.

• Remento turns spoken stories and memories into a printed memoir with QR codes to hear later.

• Firstleaf scans wine labels and menus to instantly recommend wines that match your taste preferences.

• Icebreakers.wiki generates fun ice-breaker questions for teams and group discussions (also perfect for party games).

• ulist keeps you organized by combining lists, tasks, and notes in a single app.

🤖ChatGPT Prompt Of The Day: Prompts for Brainstorming and Ideas.

Brainstorming can sometimes feel like a challenge, but with the right prompts, the process becomes smoother and more fun.

Whether you're a writer looking to create engaging content or someone exploring new ideas, these prompts are the perfect starting point to spark your creativity and fuel your next project.

See this as an AI WoD!

Tell it about yourself: “I’m a 30-something-year-old writer who wants to publish engaging content on my blog. I like reading, writing, plants, traveling, and freelancing… What are some topics I could write about?”

Ask for a huge dump of ideas: “Generate a list of 50 headlines that have to do with XYZ.”

Input the key points you want to cover and ask for an outline: “Here are the four main points of my article—generate a detailed outline that expands on them.

Source

That's all we've got for you today.

Did you like today's content? We'd love to hear from you! Please share your thoughts on our content below👇

Your feedback means a lot to us and helps improve the quality of our newsletter.

🚀 Want your daily AI workout?