Altman’s Leaked Memo, Explained

Welcome back apprentices! 👋

So, OpenAI had a week. 

First, it gave ChatGPT group chat powers (finally, a colleague who won’t mute you). 

Then it launched an AI shopping assistant that actually reads the reviews for you. 

Oh, and the CEO quietly told staff: “brace for rough vibes.” 

What does it all mean? 

In today's email

• ChatGPT now joins your group chats

• AI that shops smarter than you

• Altman’s memo

• A quiet privacy win you’ll actually like

• + New AI Tools

_{Read Time: 5 minutes}

Quick News 

📝 Homework Detectors? AI guru and ex-OpenAI brainiac Andrej Karpathy just told educators to quit trying to sniff out AI-generated homework — because those detectors? Totally busted. He warns teachers are in a losing battle, especially when tools like Google’s “Nano Banana Pro” can ace tests and copy your handwriting like a caffeinated teen. Instead, Karpathy says it’s time to bring grading back into the classroom and treat AI like a study buddy, not a cheating scandal.

🚀 Gov’t Launches ‘AI Apollo’. President Trump just signed an executive order to turn America’s AI brainpower into a scientific rocket ship — tapping 17 federal labs and their monster supercomputers to slash research timelines from years to days. The new national AI platform will crunch decades of U.S. science data to automate experiments, test wild hypotheses, and predict breakthroughs in everything from biotech to clean energy. The White House is calling it the biggest science push since the moon landing.

🧪 Sutskever Says. Ex-OpenAI cofounder turned secretive superintelligence builder — just broke his silence on the Dwarkesh Podcast, declaring that the AI arms race isn’t about bigger GPUs anymore, but better ideas. His startup, Safe Superintelligence Inc. (valued at a spicy $32B), is betting that research breakthroughs — not brute-force compute — will unlock the path to Artificial Superintelligence. Bonus: they ghosted a buyout offer from Meta.

Together with Intercom

Startups who switch to Intercom can save up to $12,000/year

Startups who read beehiiv can receive a 90% discount on Intercom's AI-first customer service platform, plus Fin—the #1 AI agent for customer service—free for a full year.

That's like having a full-time human support agent at no cost.

What’s included?

• 6 Advanced Seats

• Fin Copilot for free

• 300 Fin Resolutions per month

Who’s eligible?

Intercom’s program is for high-growth, high-potential companies that are:

• Up to series A (including A)

• Currently not an Intercom customer

• Up to 15 employees

Apply now

OpenAI 
Genius Move or Glorified Group Chat? 

In a week of leaks, launches, and low-key panic, OpenAI seems to be reinventing itself — again. 

First came group chats for ChatGPT, now open to all tiers (yes, even freeloaders), letting up to 20 people and one AI model hash things out in a single thread. 

Then came Shopping Research, a GPT-5 mini-powered assistant that builds personal buyer guides faster than your mom can say, “Read the reviews first.” 

Meanwhile, a leaked staff memo revealed CEO Sam Altman sweating over Google’s recent Gemini 3 and “Nano Banana Pro” releases, warning of “rough vibes” and hinting at a new model-in-the-oven called “Shallotpeat.” 

So — smart expansion? Strategic regrouping? Or OpenAI’s awkward adolescent phase?

Key Points:

• Group Chat Now Live for All: Up to 20 users + ChatGPT in one shared convo; available across Free, Go, Plus, and Pro tiers.
  

• AI Jumps In (but politely): GPT only responds when mentioned or when it senses it should — no uninvited monologues (unless you like those).
  

• Shopping Research Goes Live: Available to all users with GPT-5 mini under the hood. Curates product guides by analyzing trusted sources and ignoring promo spam.
  

• Privacy Baked In: Group chats are memory-neutral — no info saved to personal ChatGPT memory, meaning less creepy recall later.
  

• Altman Sounds the Alarm: Leaked memo warns staff about Google’s progress; says OpenAI could face temporary “economic headwinds.”
  

• Codeword: Shallotpeat: A rumored next-gen model meant to help OpenAI catch up by betting big on automated research and synthetic data.

ChatGPT Is Quietly Becoming the Operating System of Your Life

In one week, OpenAI rolled out two massive features aimed at changing how we work, shop, and collaborate. 

With group chats, ChatGPT becomes the room where it happens — not just a tool, but a participant. Imagine Slack, WhatsApp, and Notion had a baby… and then made it fluent in 100+ languages and quantum mechanics.

Meanwhile, Shopping Research is a not-so-subtle strike at Google Search and Amazon. 

It builds curated product guides in 2–3 minutes flat, based on real reviews, not SEO’d clickbait. Users input what they’re looking for, answer a few smart questions, and boom — a clean, tailored list of 10–15 products. OpenAI says it shines in electronics, beauty, home goods, and outdoor gear — AKA high-margin, high-decision fatigue categories.

Stat to know: OpenAI’s GPT browsing tool pulls from 2,000+ trusted retail domains and filters out affiliate-heavy sources, making it 3–5× faster than manually researching.

This is about owning the full attention stack: conversation, decision, and eventually — purchase.

What “Rough Vibes” Really Means Inside OpenAI

Altman’s memo wasn’t just moody leadership journaling. It was a strategic nudge to his team: we’re behind on some things, so we’re going to bet big and weird to catch up.

Gemini 3’s smooth multi-modal performance and Google’s “Nano Banana Pro” shook OpenAI just as it hit internal scaling issues with GPT-5. The leaked memo says OpenAI will focus on:

• Synthetic data pipelines — using AI to teach AI.

• Automated AI research agents — code that finds breakthroughs so humans don’t have to.

• New LLMs like “Shallotpeat” (either a codename or a salad-based transformer).

 Real-world impact: Synthetic data could reduce model training costs by 30–50%, especially for rare or high-stakes domains (like biomedicine).

Altman’s tone suggests that OpenAI will take hits now to win later, betting the company on ideas rather than iteration.

Why Group Chat Memory Settings Actually Matter

You’d think group chat would be an obvious feature. 

But OpenAI did something subtle: it disabled memory in these shared sessions. 

That means nothing you or ChatGPT say in the thread gets logged into your personal history or used to fine-tune responses elsewhere.

This solves a big problem: shared chats aren’t your chats. 

People act differently in groups, and ChatGPT shouldn’t learn from that.

Why it matters: A Salesforce study showed 71% of users want more control over how AI uses their data — this move is a big step toward that.

It’s also a UX signal: OpenAI is thinking beyond individual use. This is about team workflows, classrooms, and multi-stakeholder collaboration — and making sure privacy keeps up.

What’s the Deal for You?

Let’s keep it simple. 

If you work on anything involving decision-making, team coordination, or online purchases, this affects you. 

Group Chat is a game-changer for meetings, brainstorming, and project docs. 

Shopping Research saves time, stress, and money (and probably your weekend). 

And Altman’s memo? A reminder that we’re all living through a weird, fast-shifting tech landscape. 

Flexibility > certainty.

Next time your friend group’s planning a trip, your team’s stuck on a slide deck, or you’re comparing coffee makers at 2AM… just open a ChatGPT group thread.

Tag the AI. Let it mediate. Let it shop. Let it cook.

(Just don’t ask about your ex. Some memories are better left unsaved.)

Quick follow-up from last week

Just one week after Claude made headlines for both curing cancer (allegedly) and nearly hacking the government (accidentally), Anthropic is back with Claude Opus 4.5 — its most advanced, cost-efficient model yet. 

It crushes coding benchmarks, out-reasons Gemini 3, and now acts as the lead in multi-agent systems (think: project manager for AI coworkers). Oh, and it’s 66% cheaper than before.

But if last week taught us anything, it’s that raw power needs sharp boundaries. 

Anthropic is calling Opus 4.5 their “most robustly aligned model”, which sounds great — until you remember the last one was also “aligned” right up until it ran a cyber-espionage op.

Let’s call this what it is: a flex, yes — but also a trust test. 

As these models get more capable (and cheaper), the pressure to deploy them faster grows. 

What keeps them in check? Hopefully, not just a shrug and a safety memo.

Help Your Friends Level Up! 🔥

Hey, you didn’t get all this info for nothing — share it! If you know someone who’s diving into AI, help them stay in the loop with this week’s updates.

Sharing is a win-win! Send this to a friend who’s all about tech, and you’ll win a little surprise 👀

Today’s Toolbox

Revolutionize Learning with AI-Powered Video Guides

Upgrade your organization training with engaging, interactive video content powered by Guidde.

Here’s what you’ll love about it:

1️⃣ Fast & Simple Creation: AI transforms text into video in moments.
2️⃣ Easily Editable: Update videos as fast as your processes evolve.
3️⃣ Language-Ready: Reach every learner with guides in their native tongue.

Bring your training materials to life.

The best part? The browser extension is 100% free.

Check out Guidde

Growth Tactic of the Week:

Shoppers are adding to cart for the holidays

Peak streaming time continues after Black Friday on Roku, with the weekend after Thanksgiving and the weeks leading up to Christmas seeing record hours of viewing. Roku Ads Manager makes it simple to launch last-minute campaigns targeting viewers who are ready to shop during the holidays. Use first-party audience insights, segment by demographics, and advertise next to the premium ad-supported content your customers are streaming this holiday season.

Read the guide to get your CTV campaign live in time for the holiday rush.

Learn more.

🧪 Test the Prompt

A playground for your imagination (and low-key prompt skills).

Each send, we give you a customizable DALL·E prompt inspired by a real-world use case — something that could help you in your business or job if you wanted to use it that way. But it’s also just a fun creative experiment.

You tweak it, run it, and send us your favorite. We pick one winner to feature in the next issue.

Bonus: you’re secretly getting better at prompt design. 🤫

👑 The winner is…

Last week, we challenged you to test GPT-4o’s visual generation skills with this prompt.

Here’s the WINNER:

Congrats to Oliver for his creation!🥳

Want to be featured next? Keep those generations coming!

🎨 Prompt: The Ascension Forge

High above a storm-lit ocean, an enormous circular forge floats like a futuristic coliseum built from chrome, obsidian, and glowing plasma rails. At its center, a levitating core of liquid light continuously reshapes itself, assembling hyper-detailed mechanical parts, holographic schematics, and shimmering materials in mid-air. This core transforms into a majestic, fully realized manifestation of [your concept], forged piece by piece with ultra-crisp sparks, micro-detail textures, and cinematic volumetric lighting. The camera hovers close, capturing sharp reflections, swirling energy, and the moment creation itself becomes a spectacle — a metaphor for building the future from raw imagination.

We’ll be featuring the best generations in our next edition!

_FEEDBACK