When curiosity ceases to be content and becomes investigation

Information that passes, knowledge that stays

We've never had more access to information. News, papers, technical documentation, videos, debates, language models capable of summarizing hundreds of pages in seconds. Sometimes I think that the contemporary problem has ceased to be access and has become something else: the ability to transform this access into understanding.

Every day, something goes through my feed. Most of them disappear before I even finish reading the title. Some things, however, generate a little friction — that "this is interesting" that lasts a second longer than normal. Most of the time, friction also disappears. In others, it becomes a question.

This is where the routine I want to describe here begins.

I used to finish a feed reading session with the feeling of having absorbed dozens of brilliant ideas. Hours later, I couldn't explain any of them properly. The information had passed, but it hadn't. I recognized the terms. I couldn't reproduce the mechanisms.

LITERATUREThere is a plausible explanation for this pattern in information foraging theory: humans navigate in rich environments by following clues about the value of a source — what Pirolli and Card called the "aroma" of information.¹ Digital feeds are environments designed to maximize this aroma. Provocative titles, numbers, promises of novelty. The cost of extracting information is very low. The cost of digesting it remains high. The result is a continuous search behavior that rarely becomes integration.

I'm not saying the feed causes amnesia. I'm saying that it rewards hunting, not digestion — and that confusing the ease of finding something with the acquisition of knowledge is a mistake that the very architecture of these environments makes tempting.

When something breaks the flow

Once in a while, information doesn't disappear.

I can't explain exactly why. Sometimes it's a question that keeps hammering. Sometimes it's a contradiction between two sources I didn't expect to find. Sometimes it's a word I've seen ten times and never stopped to really understand. What changes is not necessarily the content — it is my relationship with it.

LITERATUREThe information gap theory describes this moment as the perception of a distance between what I know and what I would like to know.² Experimental studies associate this perception with greater willingness to seek answers and, in certain contexts, a more robust memory for the material that solves the gap.³ Curiosity, in this sense, is not only a passing affective state: it guides attention and directs the search.

But curiosity alone is not enough. She could point to an irrelevant triviality. It can be captured by a fluent and wrong source. It could last 30 seconds and disappear. What turns curiosity into investigation is a question.

Question Changes Game

The difference between consuming and investigating is not in the volume of content consumed. It's in the decision to convert an impression into objective.

When I decide I want to understand how something works — not just What? It's -- my way of engagement changes. I stop collecting loose fragments and start looking for a mechanism. It changes what I read, how I read and what I actually retain afterwards.

LITERATUREThe literature on guided research supports this distinction: the investigation conducted by questions tends to overcome, on average, free and unassisted exploitation, and guidance — criteria, prompts, some kind of support — seems to be the decisive factor.⁴ The question doesn't have to be born perfect. It can be reworked halfway. But it has to exist.

In my routine, the guide question works as a filter. Sources that do not help answer it are discarded, even when they are interesting for other reasons. This reduces the temptation to accumulate material just by accumulating. Curation begins when I accept that I do not need any more sources — I need the right sources.

Investigating went cheap. Judging remains expensive

Gathering original documentation, technical articles, websites, videos and academic references used to be the slowest part of any investigation. Today the search is almost instantaneous. What got expensive was the trial.

The technological transformation of recent decades can be described in layers: from traditional search to machine learning, from language models (LLMs) to systems that seek and cite specific sources (the so-called RAG — retrieval-augmented generation), and these to agents able to use tools and perform linked tasks through automations. Each layer reduced the cost of a specific operation: locate, compare, summarize, formulate questions, organize.

The relevant point is not chronology. It is the change of function: the machines stopped only storing and recovering information and started to participate in the manipulation, comparison and exploitation of this information. This radically alters the cost of investigating. However, it does not eliminate the need to judge.

LITERATUREThere's a risk embedded in this whole facility. The literature on cognitive offloading it shows that delegating mental processes to external devices can improve immediate performance, but reduce retention and autonomous reasoning in the long term.⁵ If the machine organizes, summarizes and explains everything for me, I lose just the friction that builds memory. The speed of the machine is not the speed of my understanding.

Machine Enters Workshop — Not Throne

I use tools like NotebookLM (or Gemini Notebook, its latest variant) as a workbench, not as an oracle.

The difference is simple to describe. An oracle answers. A bench organizes, confronts and allows questioning. I carry sources that I have previously selected — technical documentation, articles, videos, sometimes academic papers — and ask questions about them. The tool locates passages, compares different formulations, suggests relationships that I hadn't noticed. I check every relevant statement in the original source. The tool accelerates processing; judgment on quality, relevance, causality and context remains mine.

INTERPRETATIONThis is compatible with the idea of cognitive scaffolding (scaffolding): the tool assumes part of the extrinsic load — noise, dispersion, difficulty crossing dozens of sources — releasing resources for the load that really matters, that linked to the effort of integration and construction of meaning. But the metaphor has a limit that I prefer not to forget: a scaffolding does not build the building. He supports the builder.

LITERATURERecent research on generative AI in educational contexts reinforces this distinction in a not subtle way. A field experiment with high school students showed that unrestricted access to an AI assistant increased performance during practice, but reduced later performance when the tool was removed; when the use was accompanied by limits and some kind of tutoring, the damage decreased considerably.⁶ The conclusion is not "AE impairs learning." It is more precise and less comfortable: the design of use is what decides whether the tool expands or replaces cognitive work.

The Speed Paradox

After gathering and organizing the sources, I could ask the AI to generate a ready synthesis. I often do. But synthesis is not the end of the process — it is only the beginning of the hardest part.

And here I come to the question which may be the center of this text: if an AI can produce an extremely convincing explanation in a few seconds, how do I know if I really understood it?

LITERATUREThe illusion of explanatory depth is one of the most well documented phenomena of cognitive psychology. Rozenblit and Keil have shown that people often feel that they understand complex systems with much more precision and depth than they can really explain when they are tested.⁷ The feeling of familiarity is easily confused with real causal competence.

Recognizing is not the same as reproducing.

— the third distinction which this text tries to preserve from beginning to end

Reading a fluent AI-generated summary produces exactly this kind of familiarity. The text is clear. Logic seems obvious. I recognize the terms, the relationships, the vocabulary. But recognizing is not the same as reproducing.

The explanation test

LITERATUREThe literature offers a known antidote, although it is not automatic: recovering memory information, without consulting anything, produces longer-lasting retention than rereading or reviewing a text already ready.⁸ That's what research calls recovery practice. It works because the difficulty of producing a zero response strengthens memory in a way that the difficulty of just recognizing an answer does not strengthen.

EXPERIENCEThen I close the laptop. Without consulting. And I try to write or record what I understand. Which parties are involved? What state changes? By what mechanism? What would happen if the cause was absent? What does the source really show — and what am I just assuming it shows?

It is in this instant, almost always, that the illusion breaks.

LITERATURESelf-explanation seems to function as a second antidote: when I try to articulate the "how" and the "why" without external support, I usually detect inconsistencies in my own reasoning that passive reading simply did not reveal.⁹ There is also a third relevant mechanism, called in the literature of teaching expectation: studying knowing that you will need to explain the material to another person seems to organize the memory in a different way, leading to actively seek key concepts and retain for longer the most important information — an effect that appears even when real teaching never happens.¹⁰ Meta-analyses about preparing to teach and explain in fact indicate, on average, organizational gains and understanding, although the effects vary greatly according to the modality, interactivity and causal quality of the explanation produced.¹¹

EXPERIENCEI can't tell for sure if my brain encodes differently when I'm preparing a video to publish. But I understand, in a very concrete way, that the quality of my attention changes. I start looking for causal relationships. I'll look for examples. I will look for exactly the point that is not yet clear to me.

Storytelling as an instrument of thought

The script I use to structure the videos I publish insists on some elements: curiosity, a real problem, tension, progression, cause and effect relations, examples, analogies, absence of unnecessary jargon, preservation of the nuances that the original source carried. For a long time, I've treated this as a communication technique. Today I also see it as an instrument of thought.

Organizing an explanation for another person forces the reorganization of the mental representation of the person who explains it. A narrative requires sequence. It requires causality. It requires me to know what comes before and what comes after. It is simply not possible to narrate consistently a mechanism that is not understood — at least not without the seams appearing.

LITERATUREOnly there's a symmetrical risk I can't ignore. Research on narrative persuasion shows that well-built stories are usually more effective than abstract arguments to alter beliefs and attitudes.¹² The so-called narrative transport — that state of immersion in which history "takes us along" — tends to reduce counterargument.¹³ Fluency can be confused with precision with a disconcerting ease.

OPEN HYPOTHESISTherefore, the narrative, in my practice, needs brakes: carefully distinguishing what is a fact from what is an interpretation, exposing the limits of what I am saying, indicating exactly what a source shows and what remains just my hypothesis. Historytelling was born as a tool to communicate knowledge, but it can also function as an instrument to find out where knowledge is not yet well structured in the head of those who tell the story.

Make it explicit

The video posted on the channel is not the end product of any of this. It's outsourcing.

When an idea ceases to exist only in my head and comes to exist as a public artifact, it becomes observable, reviewable, questionable, reusable — and that changes the game. It's the passage from "I have a feeling I understand" to "I can turn what I understand into something that someone else can keep up with."

That's where my channel on YouTube enters history — not as a showcase, but as the place where part of this process becomes public. Publishing allows you to register the route, not just the completion. It allows other people to find the same question I asked myself. It allows to receive counterpoints that I would never have formulated alone.

LITERATUREThe literature on social accountability (social accountability) suggests that the expectation of having to justify a position for others can induce a more careful information processing, a kind of early analytical pressure.¹⁴ That's consistent with how I feel about recording and publishing. It is, however, not scientific validation of anything.

Sharing is not the same as validating

Comments, views and engagement are not evidence that an idea is correct. Popularity is not true. LITERATUREThe audience can increase the sense of responsibility, but it can also generate anxiety, self - promotion, or an excessive focus on performance rather than precision — the effect seems to depend heavily on the type of audience and context.¹⁵

What I get from publishing, then, is not approval. It's the concrete possibility that someone else will find the idea, add something I didn't see, contest a specific point, ask a question I hadn't asked.

The cycle continues

Each answer can generate a new question. Each explanation can reveal a gap. Each source can lead to the other. What I describe here is not linear — it is best understood as a cycle.

Interpretation of the practice described in this essay — not a method empirically validated as a closed package.

1. 01

   Exposure to information

2. 02

   Curiosity

3. 03

   Question

4. 04

   Research

5. 05

   Curated fonts

6. 06

   Confrontation of prospects

7. 07

   Summary

8. 08

   Reflection on own understanding

9. 09

   Explanation

10. 10

    Externalization

11. 11

    Sharing

12. 12

    Feedback → new curiosity

LITERATUREThe research on self-regulated learning describes iterative processes of planning, monitoring and adjustment that are quite similar to what I am describing,¹⁶ but no one validated, as a closed package, a public cycle that combines algorithmic feed, generative AI and video publication. What I can say, with the honesty that personal experience allows, is that every video I publish generates more questions than answers.

The cognitive ecosystem

Searchers, knowledge bases, scientific articles, LLMs, RAG systems, agents, automations, research notebooks, publishing platforms. All these tools together form something that can be described as a cognitive ecosystem — a set of external resources that actually participate in my thinking process.

The interesting question is not whether the technology "enlarges the mind". This metaphor is seductive, but too imprecise to be useful. The question is how do I structure my interaction with these tools.

If the AI organizes sources and I keep doing the trial, it broadens what I can do. If the AI organizes, judges and explains, and I just accept the result, it replaces a part of the work that should remain mine. The risk, in this sense, is not technological — it is of personal design and discipline.

What I've learned so far

I don't have a scientifically validated method to offer. I have a practice that works for me, which finds partial support — not integral — in the literature on learning, and which I keep adjusting.

What I've been learning is that curiosity alone isn't enough. That the investigation needs a question. That the curator needs criteria, not accumulation. That the synthesis generated by AI needs verification. That the explanation needs to be made without consultation, at least once, for real as a test. That the narrative needs epistemological brakes. That publication is not the end of anything — it is an opening.

Technology has really and measurablely reduced the cost of manipulating information. The responsibility to transform this manipulation into understanding remains stubbornly human.

Perhaps the question, after all, is not how many interesting information passes before us every day. Maybe that's how many of them we can turn into questions that really deserve to be investigated.