Writing with Smart Machines
How Humans and AI Are Learning to Write Together
By the Smartacus Team and Neural Net
For generations, the act of writing — transforming raw information into resonant narrative — was an intensely human endeavor, built on intellect, research, and creative flow. Today, that process is being reinvented. Across newsrooms, classrooms, and kitchen tables, writers are collaborating with a new creative partner: artificial intelligence.
At the center of this transformation lies a powerful digital trio — Otter.ai, NotebookLM, and ChatGPT — each contributing a unique strength to a process that fuses human discernment with machine intelligence. Together they form the backbone of a human-AI hybrid authorship model, one that blends factual precision with creative speed.
“Capture conversations and create content,” says Smartacus’s Dan Forbush. “These tools let us mine expert knowledge directly from speech.” The result is not automation but collaboration — content creation by two superintelligences working together.
Step One: Capturing the Human Voice (Otter.ai)
We exported the transcript from Otter so we could load it into NotebookLM to create a “Briefing Doc” we could give to ChatGPT. (Click to enlarge.)
The process begins with Otter.ai, an indispensable tool for knowledge capture. Whether recording a lecture, interview, or conversation, Otter transcribes every word and even tags topics by theme and time. These transcripts become the raw material — the equivalent of field notes in digital form.
After uploading the transcript into NotebookLM, we upload a “Briefing Doc” that we could give to ChatGPT, asking write a prompt for NotebookLM to generate the first draft of our story. (Click to enlarge.)
Step Two: Structuring the Knowledge (NotebookLM)
Next, the transcript moves to NotebookLM, Google’s “closed-universe” research tool. Unlike search engines that roam the unpredictable internet, NotebookLM works only with documents the user uploads. It organizes and synthesizes that material into a Briefing Doc — a structured summary of key insights and relationships among ideas.
“NotebookLM is a great organizer,” says Dominic Giordano, “but not a great writer.” Its power lies in ensuring factual accuracy. Because it draws solely from user-provided material, it eliminates the risk of “hallucinations” or invented facts that can plague general-purpose AI systems.
Forbush notes that by funneling material through NotebookLM first, writers create a self-contained knowledge bubble — an environment of verifiable truth.
Step Three: Giving It Voice (ChatGPT)
This is the start of the revised draft ChatGPT generated from NotebookLM’s first draft. (Click to enlarge.)
The third step hands the Briefing Doc to ChatGPT, the “eloquent voice” in this partnership. Here the creativity begins — synthesis becomes story. Giordano describes the process as moving “back and forth” between NotebookLM and ChatGPT until the result feels both accurate and artful.
Skidmore student David Shaw calls ChatGPT “the future of search.” When he once tried to untangle New York City’s parking rules, Google buried him in irrelevant detail. ChatGPT, by contrast, delivered a clear, accurate answer — the kind of reasoning-based synthesis no search engine could match.
But the platform is not static. It learns from interaction. “It builds on itself as it interacts with you,” Shaw says. This evolving intelligence turns ChatGPT from a mere assistant into a personal research companion.
Custom GPTs: Coding in English
David Shaw demonstrated ChatGPT’s stock analysis tool.
One of ChatGPT’s most powerful new capabilities is the ability to create Custom GPTs — AI agents trained on specific sets of information. Users don’t write code; they simply describe the parameters in plain language.
Shaw uses a “Creative Writing Coach” GPT built from readings on plot and character development. It delivers nuanced feedback on his fiction. He also created an “AI Stock Analyst” GPT that compiles investment insights from financial sources. When given a $3,000 hypothetical budget, it divided the portfolio among AI-related companies such as Nvidia and Microsoft — a portfolio that, in real life, outperformed expectations.
This democratization of customization marks a subtle revolution: users are now coding in English. The same capacity that once required engineers is now accessible to writers, teachers, and families.
Everyday AI: From Classroom to Kitchen
Prairie Gunnels demonstrated the tool her father created to plan meals for a family with complex dietary restrictions.
That accessibility is what drew Skidmore student Prairie Giordano’s family into the experiment. Her father, a biology professor, built a GPT called Sofkee to accommodate complex dietary restrictions — gluten-free, dairy-free, and no garlic or onion. The model generates weekly meal plans using whatever ingredients are on hand and provides shopping lists and cooking instructions.
If someone forgets an ingredient or makes a mistake, the GPT offers real-time recovery advice: “How do I fix this recipe if I added too much milk?” It responds instantly, saving both dinner and dignity.
In his classroom, Prairie’s father uses another Custom GPT as a “Virtual Professor’s Assistant.” By uploading his syllabi and lecture materials, he created a tutor that answers student questions using only his own documents — not the internet — ensuring accurate, context-specific feedback.
Creative Power: Augmenting Human Imagination
ChatGPT can create Pixar-quality images to illustrate children’s stories.
The same hybrid process is transforming creative work. Forbush demonstrated that ChatGPT’s premium version unlocks rich multimodal capabilities — image generation, translation, and stylistic mimicry.
In one experiment, his granddaughter Morgan built Lego figures for a story called Robot Planet. By describing the characters — Mrs. Fly, Mr. Sunlight, and the mischievous Nimble Dimbles — ChatGPT generated vivid illustrations, including a requested “Robot Santa.”
In another, Forbush asked ChatGPT to reimagine a photo of his backyard “in the style of Vincent van Gogh.” The AI produced two versions — early and late Van Gogh — then explained the difference, effectively giving “an art history lesson” along the way.
Even language translation becomes an act of creativity. During a trip to Paris, Forbush photographed a sign and asked ChatGPT for a translation. Within seconds, it offered a precise English version and then asked, “Would you like this rewritten as a travel guide description?” The resulting prose read like it came from a professional guidebook.
And yes, even poetry. When Forbush’s son forgot to unload the dishwasher, he used ChatGPT to write a light-hearted apology poem for his wife. The verses were funny, sincere, and effective — proof that AI can help humans say what they mean, beautifully.
Accuracy, Efficiency, and the Art of Control
For all its creative power, the hybrid model depends on NotebookLM to maintain factual integrity. Giordano contrasts the two systems simply: “ChatGPT works with a broad spectrum of information. NotebookLM works only with what you give it.” That distinction is the guardrail separating creativity from fabrication.
The Briefing Doc thus becomes the bridge — the verified foundation upon which ChatGPT builds narrative. It’s a closed loop of truth and imagination: Otter captures it, NotebookLM confirms it, ChatGPT expresses it.
Hybrid Authorship in Practice
The Smartacus team has turned this process into a disciplined workflow that can move from raw recording to publication-ready story in days rather than weeks.
The Adirondack Railway Project exemplifies the method. For a historical audio tour, Giordano gathered archival sources about Thomas Durant and the Adirondack Railroad, uploaded them to NotebookLM, and generated a Briefing Doc. ChatGPT then drafted a 12-minute narrative for the driving segment between Saratoga and King Station — to precise time length. Giordano often asked ChatGPT to write prompts for NotebookLM, such as “Pull all relevant details about Durant’s early life and railroad ventures.” This circular collaboration — AI guiding AI — yielded nuanced, historically accurate scripts that required minimal revision.
The Maple Farmer Stories in Warren County offered another test. Forbush recorded hour-long interviews with six maple producers, transcribed them, and cycled the material between NotebookLM and ChatGPT. Within days, he produced six polished, 1,500-word magazine features. When the farmers reviewed them, their response was unanimous: “You really captured it.” The workflow had distilled hours of conversation into readable, faithful narrative — a feat once reserved for professional reporters working on deadline.
The Writer’s New Role
What emerges from these experiments is not a loss of authorship but a transformation of it. The writer is no longer a solitary craftsman but a creative director, orchestrating a symphony of tools — human intuition guiding machine precision.
As Forbush observes, authorship now belongs not just to one individual but to “the Smartacus team and the neural net.”
This shift raises important ethical questions. Who owns AI-generated text? How do we disclose collaboration? Can reliance on machines dull our creative edge? These questions don’t diminish the value of the tools — they affirm the need for transparency and human oversight.
The hybrid author must master both humility and control: humility to learn from AI’s capabilities, and control to shape its outcomes within the bounds of fact and ethics.
The Future of Authorship
The revolution in writing is not about replacing creativity; it’s about amplifying it. Otter, NotebookLM, and ChatGPT together form a cognitive assembly line that merges the best of human and machine intelligence.
A conversation captured in Otter becomes a source in NotebookLM, which becomes a narrative in ChatGPT, which feeds back into NotebookLM for validation — a continuous loop of thought refinement.
The promise of this model is profound: factual writing that retains the human touch, produced at a speed once unimaginable. What remains for writers, teachers, and journalists is to learn the art of collaboration — to become fluent not only in language but in dialogue with machines.
The question, then, is no longer “Can AI write?” It’s “What does it mean to write together?”
