Notes from a workshop with the Harvard, Princeton, and Yale alumni group of Alaska, led by AI researcher and liaison Ed Caplan.
We are drowning in data but starving for insight. Yesterday, I sat down with the Harvard, Princeton, and Yale alumni group of Alaska to talk about a tool that’s quietly changing how curious people do research: Gemini Notebook. Here’s what we covered.
The curiosity paradox
For decades, the barrier to high-level computing was technical proficiency — the ability to code, or navigate complex architectures. That barrier has dissolved. The most vital skill in the AI era isn’t technical anymore. It’s curiosity.
The real problem most of us face isn’t a lack of information. It’s the inability to organize, synthesize, and trust the massive amount of research we encounter every day. We don’t need another “answer engine” spitting out unverified summaries from the open web. We need a thinking partner — a tool that respects our own curation while expanding our capacity to ask better questions.
A grounded, digital 3-ring binder
To understand Gemini Notebook (formerly NotebookLM), picture the classic academic 3-ring binder. You choose exactly which documents, PDFs, and links go inside. That’s the essence of Retrieval Augmented Generation, or “grounding.”
Unlike a standard chatbot pulling from the chaotic open web, Gemini Notebook builds a firewall around your sources. Ask a question, and the AI is restricted to the binder you’ve curated. As Ed put it in the workshop:
“Grounded in the way that… you put the sources in that you want to have the AI reference.”
Ed Caplan
The transparency is what makes this powerful. When the tool answers, it doesn’t just summarize — it cites. Clicking a citation highlights the exact passage in the source it drew from. That “source guide” is the clearest defense against hallucination we’ve seen, and it keeps a human as the final judge of what’s true. The tool can now search the open web too, but the most reliable setup still keeps the firewall on by default.
From answer engine to thinking partner
Most people treat AI like a vending machine: input a prompt, take the output. The thinking-partner model asks more of us. Take the Okmok volcano eruption of 44 BCE: ground a notebook in the right historical and scientific records, and you can explore how an eruption in Alaska, 6,000 miles away, contributed to the collapse of the Roman Republic. When the tool explains that Romans reported seeing three suns in the sky, it isn’t inventing a myth — it’s citing record-keepers like Pliny the Elder and Plutarch, right there in your sources.
Still, sitting in the driver’s seat means staying alert to the “sycophancy” trap. AI models are often tuned to be helpful first and accurate second — agreeable by design. The fix is remembering these are reasoning nets, not brains. Ed’s warning stuck with the room:
“Beware of anything that says it’s a reasoning model… it means that it is going through and reasoning as if it was a human reasoning net.”
Ed Caplan
The quality of what you get out is still tied to garbage in, garbage out. The reasoning is only as strong as the human curation behind it.
Multi-modal synthesis: podcasts and mind maps
The “Studio” side of Gemini Notebook turns static research into artifacts you can actually use — because most of us learn in more than one format:
- Briefing docs and critiques — concise summaries, or a critique that surfaces contradictions and gaps in your own logic.
- Mind maps — a visual way to see how disparate ideas, like Mediterranean climate shifts and Roman naval blockades, connect.
- Audio overviews — maybe the most striking feature: a deep-dive conversation between two AI hosts.
These aren’t just summaries — they’re interactive. Ed described the audio overview as something that “sounds like a podcast” and even lets you join the conversation. You can also connect your Google Play Books library, so you can work through an entire book by a thinker like Ethan Mollick as if it were a personal mentor. And to build your first binder, Gemini Deep Research can generate a starting source list before you take over the curating and vetting.
The “dispatcher” model
We’re moving from being behind the wheel to becoming dispatchers — orchestra leaders directing a fleet of AI agents that research on our behalf. But automation doesn’t replace domain expertise. Ed shared a cautionary story from Ford, which reportedly tried to replace a segment of its knowledge workers with AI — only to find the technology lacked the human judgment nuanced work requires, and had to re-hire the experts it let go.
The lesson: as we move toward the dispatcher model, the value of the person who knows what to dispatch — and how to verify what comes back — only goes up.
Security: your own cloud server
For anyone handling proprietary or sensitive data, confidentiality isn’t optional. Google now runs each Gemini Notebook on its own individual cloud server, which means more computing power and a firmer wall around your data. Data you upload isn’t used to train Google’s broader models.
Even so, we shared a few ground rules for staying safe:
If the data you’re about to upload showed up on the front page of the New York Times tomorrow, would you or your organization be in trouble? If yes, leave it out.
The front-page test
- Anonymize — swap in placeholders for names, clinical data, or other identifiers.
- Check your policy — confirm your organization’s stance on third-party cloud tools before you upload proprietary research.
The orchestra leader’s final thought
We’re graduating from the era of the instrument player to the era of the orchestra leader. Gemini Notebook isn’t a replacement for human thinking — it’s a force multiplier for human curiosity. It lets us step back from the manual labor of retrieval and focus on the higher-level work of orchestrating ideas.
The technology is ready. The question is whether you’re ready to stop being the instrument player and start being the orchestra leader. Who are your “five smartest friends” — human mentors, or grounded AI thinking partners — and how will you use them to learn faster than the rest of the world?
Leave a Reply