Ahoy, AInauts,

Welcome to a new issue of your favorite newsletter.

Today, we finally have some AI news from Germany again. Plus, a cool free tool we currently love and a rather sad AI rumor, if it turns out to be true.

Here is what we have for you:

Let's go!

Claude can be your analyst, editor, and strategist.

But most professionals are using it to fix grammar.

These 1,000+ Claude prompts take it from grammar tool to your most powerful AI work assistant.

Sign up for Superhuman AI and get:

Claim your free prompts

Normally, we don't like writing about models and tools you can't try in full right away.

But we're making an exception today.

Models like Google's Nano Banana, Veo, Seedance, and GPT Image 2.0 have dominated images and video over the past few months. America, China, America, China.

Finally, there is news from Germany again.

Black Forest Labs, based in Freiburg im Breisgau, has unveiled FLUX 3.

If the name doesn't ring a bell: some of the people behind it were part of the team that built the architecture behind Stable Diffusion.

FLUX models rank among the best image models out there, and BFL was most recently valued at $3.25 billion.

Back when FLUX 2 came out, we wrote a lot about it. Then things went quiet for a while. But it's finally here. FLUX 3.

That's the real point: FLUX 3 was trained across images, video, and audio. The same model can:

As office dwellers, that last point matters least to us personally. But as Europeans, it may be the most important one: having models of our own that can control robots.

In the first comparisons, FLUX 3 was preferred over Luma Ray 3.2 in 93 percent of cases, and over Runway Gen-4.5 in 77 percent.

Important: These are the creators' numbers and, according to BFL itself, preliminary results. We haven't tested it ourselves yet.

You can't use it yet.

Right now, only the video and robotics variants are available, and both are limited to closed early access that requires an application.

The image model, the part most of us care about in everyday work, is coming "in the next few weeks."

Still, we want to spotlight AI news from Germany. If FLUX 3 interests you, you can join the waitlist here:

→ Join the FLUX 3 waitlist

We signed up and will let you know once we're actually in and have put it through its paces ourselves.

The president of OpenAI. The CEO of Google. The founder of LinkedIn. We talk to the people building AI. Every morning, we put what they told us in your inbox.

What they're actually worried about. What they're betting on. What's coming next. Delivered in five minutes, before your first meeting.

With one email, and only five minutes, you can stay ahead of 99% of the world.

Join 75K Readers →

Now for something you can probably use right away.

Maybe you already use a whiteboard tool like Miro.

If not, consider giving one a try. They are especially useful for team collaboration.

They also help when you need to turn another wall of AI text into something visual and structured, whether it is research, a meeting summary, or an analysis.

AI often spits out a whole novel. And most of us simply understand things better visually.

That's why we're writing about a tool that keeps impressing us: tldraw.

It's a completely free online whiteboard tool with all the usual features: text boxes, Post-its, image embeds, and so on. Ideal for remote work.

There is now a local desktop app too. Free, no account, no server, and fully offline.

And here's the part that made us write about it at all:

Your AI agent can work on it with you.

We don't mean "AI draws you a picture." You and the agent are looking at the same surface for the first time. It reads the canvas, understands what's on it, rearranges things, and draws connections.

We dropped all of our FLUX 3 research from above, a nearly 4,000-character wall of text, onto the canvas and said: “Turn this into an overview for me.”

tldraw isn't just a canvas. A .tldraw file can contain code that runs every time you open it. The whiteboard becomes a small app.

One small example: The AI Leverage Board.

You enter your tasks. A click on a card cycles its status from manual to semi-automated to automated.

The summary on the left recalculates instantly and tells you where your biggest untapped lever lies.

Small, dynamic, smart apps. No account, no subscription, no server. It runs locally on your machine.

As mentioned, it's free, and you can use it directly online here.

But we also recommend downloading the desktop app for a 100% local setup.

If you use Claude Code, Codex, or something similar, you can install the agent skills with one click. Then your AI employee can build cool whiteboards with you.

The big difference from Miro and FigJam: the behavior lives in your file, not in a subscription. Try it and let us know whether it helps!

Let's wrap up with a topic that went viral this week.

Quick question to set the scene: What actually makes today's AIs as good as they are?

Sure, they ingest the whole internet. BUT: The really good knowledge is in books. Especially old books.

Reports are circulating that AI companies are buying rare, old, and out-of-print books to use for training their models.

Apparently, something pretty ugly is happening in the process. To make the books easier to scan, their bindings are cut off and the pages separated. After scanning, everything is apparently destroyed. In some cases, these are centuries-old books with few copies left.

Right now, it seems antiquarian booksellers in Germany, the Netherlands, and Spain are being flooded with bulk orders. One dealer went from 20 to several hundred books a week. But no one knows who's buying. The buyers remain anonymous.

If that's true (and we don't know that 100%), that would be pretty sad.

The race for knowledge is real, though. One recent case fits right in: beginning in 2024, Anthropic bought millions of used books under the internal name "Project Panama," cut off the bindings, scanned the pages, and disposed of the remains.

In an internal document, it says apparently verbatim: "Project Panama is our effort to destructively scan all the books in the world." And right next to it: "We don't want it to be known that we are working on this."

Before that, Anthropic had tried using around 7 million books from shadow libraries. That ended with a $1.5 billion settlement, finalized on Monday.

Books published before 2022 are premium. Because they definitely don't contain AI-generated text.

Human-made work is becoming a scarce resource. Your old archive, your own notes, your own style suddenly have value.

We expect a wave of hype around 100% human-made work.

We're sticking with AI anyway, and we want it to keep getting better. That's obvious. But it's worth knowing what all of this is built on.

One last thought that we'll also explore in more depth:

For the past few years, we've tried to hit 10,000 steps a day, exercise, and so on to compensate for all that sitting. Going forward, it may become more important to write down our own thoughts without AI's help, simply to keep our brains and ability to think in shape. 😁

So, that's all for today. As always, a huge thank-you for reading.

See you next week.

Reto & Fabian from AInauten