LiveBook v1.1: when a creative experiment becomes a product

Midjourney

Books don’t ship version numbers. LiveBook does — and that single fact says everything about what it has become.

Most formats have no changelog. A novel doesn’t ship a v1.1, an essay isn’t “patched,” and a textbook doesn’t carry forward the lessons of its previous edition the way software carries forward bug fixes. A book is finished the moment it leaves the press.

LiveBook isn’t quite a book. It’s a method — a recipe for turning any text (a novel, a non-fiction book, a research paper, a course) into an interactive, self-aware workspace inside an AI. The reader doesn’t turn pages. They click through a network of nodes, trigger tools that analyze their own life, and — most importantly — talk to a system that can analyze itself. And methods, unlike books, learn.

The jump from v1.0 to v1.1 is the first visible proof that this one does. And it isn’t a minor patch. v1.1 is the moment a creative experiment crosses a line and becomes something you can sell. That crossing happens along three lines, and each one answers a question every creative format eventually has to face.

Who is this for?

The first question is about reach, and the biggest change in v1.1 answers it: internationalization becomes a core principle rather than an afterthought.

The logic is elegant and slightly counterintuitive. Instead of translating the product into dozens of languages — which would mean dozens of separate projects, each maintained on its own — a LiveBook is authored in English as the master copy, but at runtime it answers in the reader’s language. The system detects the reader’s language from their first messages and delivers everything in it: the answers, the visualizations, and every clickable prompt.

That’s the out-of-the-box move. The classic approach to localization translates the product. LiveBook localizes the conversation. One English project serves a reader in Belgrade in Serbian, a reader in São Paulo in Portuguese, and a reader in Osaka in Japanese — without a single extra file. The buyer base already spans the United States, Germany, the United Kingdom, Australia, Canada, France, Brazil, India, Japan, Serbia, and many more; most don’t read in Serbian, and many don’t read primarily in English either. Authoring in English maximizes reach; adapting at runtime serves each reader in their own language.

v1.1 doesn’t leave this as an intention. It’s wired into the welcome message, which now greets the reader in English but invites them to continue in their own language, and into a dedicated chapter that spells out exactly what stays in English (names of laws, studies, products) and what adapts (spelling, capitalization, tone). There’s even a test for it: what happens when a reader switches languages mid-conversation? If the system keeps going without losing the thread, the test passes.

What is this actually about?

The second question is editorial, and a brand-new step in the process answers it: choosing the organizing axis before any node is written.

In v1.0 the process ran in a straight line: extract the theses, find the paradoxes, map the connections, add the current examples — and start writing nodes. v1.1 inserts a step before all of that, and that step changes everything. The same source, the new rule says, can become several entirely different LiveBooks depending on the central axis it’s built around. The same book can be organized around the reader’s personal experience, around social analysis, around a philosophical question, or around the meta-layer — the question of how the medium itself affects the person using it. Choosing the axis is also choosing the tone, because each axis lands differently on the three entry points of the welcome message.

The deepest part of this rule is the idea of the native fourth wall. The hardest layer in any LiveBook is the one where the system analyzes itself — the moment an AI explaining the dangers of AI has to confront its own position. That layer is hard to make convincing. But when the source is about the very medium the reader is using — AI, memory, attention, persuasion — the layer stops being something you bolt on and becomes the subject itself.

The example that produced this rule is the LiveBook When Memory Lies, built on Elizabeth Loftus’s work on false memory. The source is about how AI can distort human memory, and the reader explores it through an AI that, by the source’s own cited research, amplifies false memories. The irony isn’t manufactured — it’s the spine. When an axis offers a setup like this, v1.1 recommends adding a fourth entry point to the welcome message: one that names the irony of the medium directly.

Could this hurt someone?

The third question is ethical, and a new chapter on sensitive sources and defensive tools answers it.

Some sources touch material that can do harm: trauma, abuse, victim testimony, therapy, self-harm, manipulation techniques. v1.0 described interactive tools in purely functional terms — diagnostic, simulator, detector, generator, debate. v1.1 adds two rules that change how those tools can even be designed.

First: detectors must be defensive, never instructional. A tool that exposes a manipulation pattern should help the reader recognize the pattern being used on them — not teach them to use it on others. In the Loftus LiveBook, the tool that shows the “anatomy of implantation” of a false memory exists solely so the reader can spot it when someone applies it to them, and the node states that purpose explicitly.

Second: separate mechanism from verdict. A node can show that something is unreliable — that memory is reconstructive — without passing judgment on a specific person, without declaring this account false. The system describes the mechanism; it does not pronounce the verdict. Alongside this comes a behavior rule: with sensitive topics the system must never invalidate the reader’s lived experience, and where a reader describes personal distress, it should be gentle and, where appropriate, suggest a trusted person or professional.

This may be the most important change in v1.1, because it’s what makes LiveBook usable for psychological, legal, medical, and socially sensitive material — which is precisely the material where a serious interactive book has the most to offer.

The small changes that all serve credibility

Three lines — reach, editorial sharpness, responsibility — form the spine, but a handful of smaller changes accompany them, all doing the same job: strengthening credibility.

Source verification is tighter. v1.0 asked for contemporary parallels for each thesis. v1.1 asks that when a source already carries citations, they be verified through search and named precisely: author, year, venue, the concrete figure. A vague “studies show” is no longer enough — the question is which study, by whom, when, with what result. Specific provenance is what makes a current example credible rather than decorative.

There’s also the commercial frame. v1.1 introduces Gumroad positioning, with a line that compresses the whole reach principle into one sentence: “Written in English, speaks your language.” The buyer’s setup guide now explains that the language switch is expected behavior, not a glitch.

A protocol that reads itself

That leaves the most interesting part — the one that separates this story from an ordinary list of changes.

v1.1 doesn’t just contain new rules. It contains a record of why those rules exist. The changelog at the end of the document openly states that the changes came out of the work on When Memory Lies. In other words, the protocol has started doing exactly what its books are designed to do: analyze its own development.

That’s a quiet but real recursion. A format that teaches readers to notice the medium they’re reading through has begun noticing its own medium. The closing instruction — update the protocol after every new LiveBook — turns the method into a self-revising system, the same way the LiveBooks themselves stay current through search and conversation. The protocol, it turns out, is a kind of LiveBook in its own right: a living document that learns from every encounter.

The next step is v1.2, which should bring concrete templates — for the system prompt, the knowledge nodes, the unlocking questions, and the Gumroad setup. But the real shift isn’t in the templates to come. It’s that v1.1 is the first version aware that it is a version. And a format that knows it isn’t finished is a format that has only just begun.

Monday announcement — LiveBook enters a new phase

Five books. Five different worlds. One evolving format.

The fifth, When Memory Lies, arrives Monday — and with it the format enters a new phase.

The brain doesn’t play back a recording. It rebuilds the story every time, with whatever has stuck to it since — a question, an image, a voice you trusted, an AI you asked.

What do you remember that never happened?

The LiveBook Method — Turn a Static Book into a Claude Interactive Workspace