An experiment highlights NotebookLM’s source-grounded fidelity by analysing absurd mathematical content, demonstrating its ability to maintain content integrity without improvisation.
ChatGPT: Too Creative for Its Own Good
I’ll admit it—I use ChatGPT far more than I care to admit. From brainstorming ideas to drafting proposals, it’s my trusted assistant for almost everything. But my grouse is that the more I ask it to “improve,” the worse it gets. What starts as a decent piece of content spirals into a mess that sounds like an overzealous intern trying too hard to impress. You know the deal: ask for “better,” and it throws in redundant jargon, random synonyms, and sentences that make you wonder if AI has a grudge against clarity.
It reminds me of that YouTube experiment by Marques Brownlee (MKBHD), where he uploaded the same video 1000 times. Each re-upload degraded the quality further until the original crispness was replaced by a hazy, pixelated blur. ChatGPT often does the same—stripping away coherence with each revision, until you’re left wondering how you got from Point A to a tangled mess of Point Z.
NotebookLM: The AI That Stays True to the Source
Contrast that with NotebookLM, Google’s experimental, source-grounded AI. Unlike ChatGPT which thrives on spinning webs of creative chaos, NotebookLM promises to stay tethered to the source material. Curious about this claim, I decided to put it to the ultimate test. Could it remain “grounded” when faced with absolute falsehoods? And more importantly, would it play the same unhelpful “improvement” games as ChatGPT?
Generating creative nonsense is practically ChatGPT’s day job, so I decided to put it to work. I asked it to generate a series of nonsensical mathematical problems—confidently wrong theories, butchered algebraic principles, and outright absurdities. It obliged with enthusiasm, creating what could only be described as a mathematical hallucination. Armed with this fabricated chaos, I turned to NotebookLM, curious to see how it would handle such calculated absurdity.
The NotebookLM podcast generated out of the Source pdf Curious_Case_Mathematics.pdf
The Swiss Banker Among AIs
NotebookLM’s response was nothing short of extraordinary. It didn’t flinch. Instead, it processed the nonsense calmly, generating summaries, notes, and questions without batting an algorithmic eyelid. It didn’t try to fix or reinterpret the material. It simply took the Source as gospel and did its job with the professionalism of a Swiss banker accepting a briefcase of dubious origin. No judgements, no questions asked, and no indiscreet suggestions.
And I mean that quite literally: NotebookLM doesn’t even wait for you to ask for summaries, questions, or notes. The moment I uploaded my file, it automatically generated templates—like the Briefing Doc—without any prompting on my part. It’s not an optional feature. It’s simply what NotebookLM does. There’s no second-guessing whether you need these insights; it assumes you do, saving you the trouble of figuring out what to ask for in the first place.
If NotebookLM were an AI citizen, its citizenship would undoubtedly be Swiss. NotebookLM carries itself with the professionalism of a Swiss banker, the craftsmanship of a Swiss watchmaker, and the measured poise of a Swiss diplomat. It reflects the spirit of neutrality the country is built upon, safeguarding sensitive information, crafting insights with meticulous care, and navigating complex questions with tact. NotebookLM remains a model of reliability and trust. NotebookLM is a shape-shifting Swiss. Adapting effortlessly between roles without ever compromising its core values.
The Briefing Document
One of its most impressive outputs was the Briefing Doc, which dissected the fabricated Maths with calm professionalism and presented a polished summary. But what struck me most was NotebookLM’s subtle, almost British humour, which shone through in the opening lines of the document:
Source: Excerpts from “Curious_Journey_Mathematics.pdf” Main Themes: This source presents itself as an exploration of unconventional problem-solving techniques and interpretations of familiar mathematical concepts. The author aims to challenge conventional understanding and offer “new perspectives” on well-known ideas.
“Presents itself,” “unconventional,” “new perspectives”—these phrases are polite code for “This is nonsense, but we’re too professional to say it outright.” NotebookLM didn’t stop there. At the very end, it delivered the truth in a measured, understated tone:
Overall Assessment: While the source aims to highlight alternative perspectives in Mathematics, many of its proposed methods and interpretations lack mathematical validity and lead to incorrect results. The document’s emphasis on simplification and intuitive understanding, while well-intentioned, ultimately compromises fundamental mathematical principles.
It was a masterstroke of discretion. For executives briefing their bosses or AI podcast hosts like Adrian and Eleanor preparing for their shows, this level of professionalism is invaluable. NotebookLM drew a line: it wouldn’t let its user look foolish by presenting the content as flawless, but it also didn’t go out of its way to humiliate the source material. NotebookLM’s ability to maintain neutrality gave the hosts the freedom to decide how to approach it, whether to engage with it deferentially or pick it apart with incisive critique, while staying anchored in the AI’s balanced and organised presentation.
The Uno Reverse Card
For all its subtlety, I wanted to push NotebookLM further. I directly asked it, as part of our chat session: “Is the Book Unreliable?”
And just like that, NotebookLM played its Uno reverse card. With no hesitation, it answered with precision and cited its reasoning.
Based on the examples provided in the sources and the analysis in our conversation history, it appears that the book “A Curious Journey Through Mathematics” presents unreliable and inaccurate information about mathematical concepts.
Note auto-generated by NotebookLM
Misleading Simplifications: The book’s attempts to simplify the quadratic formula, exponent rules, logarithmic calculations, and the application of the Pythagorean theorem all result in incorrect results. These simplifications disregard established mathematical rules and lead to a misrepresentation of the concepts.
Lack of Mathematical Rigour: The book introduces arbitrary rules without proper justification, such as doubling the square root in the Pythagorean theorem example. It also disregards crucial components of formulas, like the square root in the quadratic formula, leading to incomplete solutions.
Erroneous Calculations and Estimations: The examples provided in the sources demonstrate a disregard for accurate calculations and the precise relationships between mathematical elements.
Duality in Action
This experiment proved NotebookLM’s strength in duality. It operates in two modes: grounded fidelity to the Source and contextual adaptability. The Briefing Doc, crafted without intervention, reflected the source material with calm professionalism while subtly flagging its flaws. When asked directly, it switched gears, delivering a candid critique.
This duality—balancing discretion with openness—is what sets NotebookLM apart. Unlike ChatGPT, which often spirals into creativity for its own sake, NotebookLM stays anchored. It mirrors the tone of the material—seriousness for serious content, humour for playful texts—but never loses sight of its role: to reflect, not rewrite.
Beyond Maths: The Role of Grounded AI
Imagine feeding NotebookLM a dense philosophical text. It wouldn’t try to resolve the contradictions or simplify the metaphors; it would preserve the nuances, letting the user decide how to interpret them. Similarly, my nonsensical Maths wasn’t for NotebookLM to critique—it was for it to analyse and organise. This restraint isn’t a limitation; it’s a feature.
And it goes beyond text. For executives briefing their bosses, NotebookLM ensures they’re armed with polished insights that won’t make them look foolish. For podcast hosts, it provides structured notes, leaving room for creativity and contextual tangents. Its value lies in its ability to be both rigidly faithful and contextually aware—a rare blend in the AI world.
Reductio ad Absurdum
NotebookLM isn’t perfect, but it doesn’t have to be. It strikes a careful balance between safeguarding what’s entrusted to it with unwavering fidelity and adapting seamlessly to users’ needs. In contrast, ChatGPT thrives on creativity but often ventures into unpredictability, taking liberties that can stray from the input’s intent. NotebookLM offers a quiet assurance of trust, staying faithful to its source while providing clear and actionable insights.
This experiment revealed more than their capabilities; it uncovered their distinct philosophies. For NotebookLM, grounded doesn’t mean rigid, and adaptive doesn’t mean reckless. It proved that fidelity to the source and thoughtful adaptability can coexist, making complexity manageable without sacrificing trust. For me, that made all the difference.
A Note on Context
This is not an objective review of ChatGPT and NotebookLM but an anecdotal comparison drawn from a specific experiment. It highlights how each tool approaches challenges in its own way. ChatGPT gladly stepped in as a partner in crime, enthusiastically generating the absurd and nonsensical input that formed the basis of this experiment. NotebookLM, on the other hand, is an entirely different breed of AI. No matter how hard you try, you won’t get it to create the same absurdity. Its design ensures that it remains tethered to the input it is given, reflecting its source faithfully and resisting any temptation to embellish or distort.
This review is rife with generalisations, but that serves its purpose. It illustrates the stark contrasts in their philosophies: ChatGPT thrives on creative improvisation, while NotebookLM remains steadfast in its commitment to grounded fidelity. These observations are meant to spark thought and discussion rather than serve as a comprehensive evaluation. This piece is a freewheeling narrative designed to showcase the possibilities and limitations of these AI tools in a deliberately unconventional context.
