Blog

Your information diet is making you average

There's a reason most knowledge workers arrive at the same conclusions at the same time. They're reading the same things. The trending page, the algorithm-surfaced posts, the threads that survived the group chat filter. Whatever's popular right now, this week, in your particular corner of the internet.

This is comfortable. It's also how you end up producing work that's indistinguishable from everyone else's. If a hundred people read the same article and draw conclusions from it, those conclusions converge. They have to. The inputs are the same. The only way to think differently is to read differently.

The trending trap

Most people's reading is determined by recency. What was published this week, what's being discussed right now, what the algorithm decided you should see today. This creates an information monoculture: everyone in a given field or community is absorbing the same material at the same time, which means the ideas generated from that material are also convergent.

The problem isn't that trending material is bad. Much of it is good. The problem is that it's shared. Ideas derived from widely-read material are commodities by definition. If you and five thousand other people read the same paper this morning and had roughly similar reactions, the insight you derived from it is worth approximately nothing because it's not scarce.

Original thinking requires different inputs. Not contrarian inputs for the sake of being contrarian, but inputs that other people in your field aren't consuming, which means you need to look in places that aren't being surfaced by whatever recommendation engine is feeding you.

Old material is criminally underpriced

The most undervalued information source in most fields is old material. Not old in the sense of outdated, but old in the sense of having been written before the current conversation started and now sitting outside the window of what anyone is paying attention to.

Claude Shannon gave a talk on creative thinking in 1952 that contains more useful technique per paragraph than most modern creativity books. His opening move was elegant: shrink a problem until it's nearly trivial, solve the small version, then reintroduce complexity one piece at a time. That technique alone will get you through more walls than any productivity advice published this decade.

Rich Sutton's "The Bitter Lesson" is roughly a thousand words long, written in 2019, and predicts the shape of the AI field better than surveys ten times its length. Most fields have equivalent texts: short, old, and devastatingly right. The problem is that nobody shares them because they were published before the current discourse started.

The commonplace book tradition, which goes back to Renaissance scholars who collected the best passages from their reading in notebooks they kept for life, was built on exactly this insight: the most valuable material compounds across years. A passage you captured five years ago can illuminate a problem you're facing today in ways that this morning's preprint cannot, because it comes from a different context and carries different assumptions.

If your information diet contains nothing older than last year, you're seeing a thin slice of what's available and mistaking it for the whole picture.

Range beats depth (eventually)

There's a persistent belief that the way to become an expert is to go deeper and deeper into an increasingly narrow domain. For a while, this is exactly right. You need depth to understand a field well enough to contribute to it. But past a certain point, further depth produces diminishing returns, and breadth starts to pay off in ways that depth alone cannot.

The most interesting work tends to happen at the edges between fields, because that's where ideas from one domain illuminate problems in another. Interpretability research borrows heavily from neuroscience. Evaluation design in AI is mechanism design wearing a lab coat. A working understanding of how hardware actually moves data tells you which software architecture ideas are viable before anyone runs the benchmarks.

This isn't about becoming a dilettante. It's about building enough fluency in adjacent fields to spot the connections that specialists within those fields can't see because they're too close. An hour spent reading an introductory paper in an adjacent field can produce more original thinking than ten hours reading further into your own.

The practical version of this: when you're stuck on a problem, don't read more of the same literature. Read something from a completely different field and see whether the framing suggests anything. More often than you'd expect, it does.

Read the source, not the summary

This is worth saying even though it sounds obvious: read the paper itself, not the thread summarising it.

Threads, summaries, and review posts are useful for discovering what exists. They're poor substitutes for engaging with the actual argument. The summary strips away the nuance, the caveats, the methodological details, and the limitations section, which is usually the most honest paragraph in any paper.

The appendix is where the bodies are buried: the details that didn't fit the narrative of the main text, the edge cases, the failed attempts. These are often more valuable for your own work than the headline results, because they tell you what the authors tried that didn't work, which saves you from trying the same things.

Annotating as you read changes the relationship between you and the text. When you're marking passages, writing questions in the margins, and noting connections to other things you've read, you're processing the material rather than just consuming it. The difference in retention and insight is significant.

When you're done, write a brief literature note in your own words: what was the argument, what's relevant to your work, what questions does it raise. Fifteen minutes per paper. These notes become your actual research library, searchable and linkable, and infinitely more useful than a folder of PDFs with scattered highlights.

Build a library that grows with you

The information diet isn't just about what you read. It's about what you keep and how you find it later.

A research library that accumulates your reading, your notes, and your reactions over months and years becomes a resource that no amount of re-reading can match. When you're working on a new problem and search your library for related concepts, you'll find things you'd forgotten you'd read, connections between ideas from different periods of your reading, and your own past thinking on related topics that saves you from repeating the same work.

Semantic search makes this practical even without perfect filing. You don't need to remember what you called something or which folder you put it in. You describe what you're looking for and the system finds it based on meaning. "That thing about attention mechanisms and energy regulation" surfaces the relevant papers even if you filed them under different project names at different times.

The AI assistant can synthesise across your collected reading: "What have I read that relates to the relationship between X and Y?" produces an answer grounded in your own library rather than a generic summary from the internet. This is the difference between an information diet and an information infrastructure.

The practical version

If you want to change your information diet without a dramatic overhaul:

Add one non-obvious source to your regular reading. A journal from an adjacent field. A blog by someone with a different methodological tradition. A book written before 2000 on a topic you care about. One source is enough to start introducing variation.

Spend an hour with a classic text in your field that you've been meaning to read but never have. The text that everyone cites but nobody seems to have actually read is usually the one worth reading.

When you find something that changes how you think, capture it properly: clip it, write a note on why it matters, link it to related things you've read before. The reading only compounds if you keep it somewhere you can find and build on.

Cut one source that you read purely because everyone else does. You probably already know which one it is.

Frequently asked questions

How do I find reading material outside my usual sources?

Follow references in papers you've already read, especially older references that nobody seems to cite anymore. Ask people in adjacent fields what they consider foundational. Browse the reading lists of researchers you admire. Attend talks or read papers from a field you know nothing about, once a month. The unfamiliar is where original inputs live.

Won't I fall behind if I stop reading the latest publications?

You won't stop entirely, you'll still encounter the important recent work through colleagues, conferences, and targeted searches. What changes is the ratio: instead of 100% current and trending, you shift to maybe 60% current and 40% older, broader, or adjacent material. The current work still gets read. It just stops being the only thing.

How do I remember what I've read?

Write about it. A brief note in your own words after finishing a paper or chapter is the single most effective retention technique. Progressive summarisation works too: highlight on first read, bold the key highlights on second pass. But nothing replaces writing your own reaction, because the act of formulating your response forces processing that passive reading doesn't.

How many sources should I be reading regularly?

Quality matters far more than quantity. Three papers read carefully with literature notes and genuine engagement will produce more insight than twenty papers skimmed and bookmarked. If your reading list is growing faster than your understanding, you're collecting rather than learning. The collector's fallacy is one of the most common traps in knowledge work.