The Default State of Information Is Noise
The previous essay laid out the framework: information, records, knowledge, and creation are four distinct stages. This one deals with only the first — information.
The single most important thing to understand about information is this: its default state is noise.
We have a natural greed for information. See a good article, want to bookmark it. See a great quote, want to save it. See a reading list someone recommended, want to add it to your queue. Every act of “saving” gives us a small illusion of “I’m making progress.” But saving, at its core, is just moving information from someone else’s warehouse into yours — it hasn’t gone through any processing by your own machine.
So the work of the first layer isn’t storing more — it’s blocking more.
The test for whether a piece of information should be let in isn’t “is it useful” — almost all information is “potentially useful,” and that’s exactly the root cause of an exploding bookmark folder. The test should be: is it worth moving into the next stage of processing? In other words, are you willing to spend time recording it, structuring it? If not, then it’s noise to you, no matter how “useful” it looks — and it should be kept out.
Splitting Information Into Three Categories
Talking about “information” in general isn’t actionable. I split incoming information into three categories and treat each differently.
Category one: what to show AI. This is raw material for the model — things you plan to have AI retrieve, summarize, compare, or process. Industry reports, documentation, other people’s long-form writing, your own historical records all belong here. They don’t need to be read word by word, or even read right now — they just need to be clearly structured enough for the model to call on. For this category, you can capture freely, because it’s AI processing it, not your attention.
Category two: what to skim casually yourself. This is low-value input that keeps your instincts warm and your sense of the environment current: industry moves, what your peers are doing, what’s trending on a platform. Each individual item has low value, but skimming them in bulk gives you a feel for “right now.” For this category, the key is limiting the time and the entry point — it’s the useful tier of background noise, but it must never be allowed to occupy your deep-focus hours.
Category three: pure noise. Emotionally charged hot takes, opinions written just to pick a fight, an endlessly scrolling recommendation feed. Their only effect is consuming your attention. For this category, there’s exactly one move: block it, without mercy.
The ratio of these three categories determines the health of your information layer. Most people’s problem is: they skim category three as if it were category two, they save category two as if it were category one, and eventually category one — the raw material that should actually be fed systematically to AI — gets buried under everything else, unused.
In the AI Era, the Signal-to-Noise Ratio Is Deteriorating
Someone might say: information overload is an old problem, what does it have to do with AI?
A lot, actually. AI has upgraded information overload into “synthetic information overload.”
Old noise was at least written by humans, and production had a ceiling. Now, models can mass-produce content that “looks very useful” — neatly structured, argument complete, quotable lines everywhere — but with no firsthand experience, no real friction, no actual information gain. It’s harder to identify than human-written noise, because it looks exactly like signal.
This leads to a counterintuitive conclusion: in the AI era, what’s genuinely scarce is no longer information — it’s signal. When anyone can generate “content” with one click, the things that can’t be generated — the pitfalls you personally fell into, the unwritten unspoken rules of an industry, specific frontline data, a trend nobody has noticed yet — become extremely valuable precisely because of that.
Someone in the community wrote a line I strongly agree with: scarcity is a precious form of wealth. Applied to the information layer, this means: don’t compete on volume where AI can generate content with one click; spend your capturing effort hunting for scarce signal instead. Scroll through fewer secondhand opinions, accumulate more firsthand experience; save less of what “looks useful,” find more of what “exists nowhere else.”
The Key to Noise Reduction Is Controlling the Entry Point
Once you understand you need to block, the question becomes: how do you actually block it?
Willpower alone won’t do it. Effective noise reduction comes from attacking the entry point, turning the arrival of information into something structured and rhythmic, rather than being interrupted at random by push notifications.
I saw a particularly plain but effective approach in that community, worth copying. The person first did one thing: audited every one of their input sources. Once counted, it turned out to be a fixed, small handful — one person’s daily short video, two livestreams a week, articles from one community. That was it. Information anxiety is, a lot of the time, an illusion: you think you need to follow a hundred sources, but in reality only three to five actually give you signal — the rest is noise pretending to matter.
After the audit, they did a second thing: moved all learning into a fixed, low-energy time slot. Before that, their learning was completely fragmented — the moment the community owner posted, they’d check it right away; scrolling their phone, anything useful got saved instantly — and the result was a day spent busy, only to look back and realize the most important work never got done. Behind this is a mindset of “fear of falling behind, fear of missing out,” and that mindset fills every single day with information you can never finish.
Once they moved input to the afternoon — their lowest-energy time of day — a pattern immediately emerged: high-energy hours were reserved for recording and creation, low-energy hours were used to “go through information.” Information stopped being a random interruption exploding at any moment, and became one scheduled stage in the day.
These two moves — auditing sources, batching processing — are the foundation of noise reduction. They don’t look like advanced techniques, but it’s precisely this kind of plain entry-point management that determines whether your information layer is clean or clogged.
Let AI Stand at the Information Layer
Finally, back to the rule that runs through this entire series: at the information layer, use AI boldly.
Capturing, deduplicating, summarizing, translating, initial classification, compressing a pile of material into a one-page bullet list — all of this is information-layer work. Hand it to AI, and you get a tenfold efficiency gain with no real loss. You don’t need to personally read every single piece — let the model go through it first, flag “the ones worth your hands-on processing,” and spend your attention only on the signal it surfaces. This is exactly where AI belongs in this pipeline.
But hold the boundary: AI helps you process noise, it doesn’t form judgment for you. It can tell you “what these ten articles say,” but it can’t decide for you “what this means to you” — that’s the job of the next stage (records), and that step has to go through your own hands. Use AI as aggressively as you want at the information layer, but the moment it comes to “what judgment should this piece of information settle into,” turn the automation off and do it yourself.
Get the information layer right, and you’ll feel lighter immediately: your bookmark folder stops carrying guilt, because you’ve finally admitted most of it is noise; your attention gets freed up, because you only spend effort on signal.
Next, the signal you’ve filtered out — worth processing by hand — moves into the second stage: records. That’s the highest-conversion semi-finished product between information and knowledge. See you in the next essay.
This is the second essay in the “From Information to Creation” column. Previous: Overview — Information, Records, Knowledge, Creation . Next: the second layer — Records.
Responses