Compression Lens

Expert explanations are compressed patterns — learn to decompress them.

Beginner6 min read

Overview

The Compression Lens recognises that experts explain complex ideas quickly because they have compressed years of experience into short mental shortcuts. Your job as a learner is to decompress: unpack what pattern sits underneath the brief explanation.

Why it matters

Beginners often feel stupid when an expert says something in one sentence that takes them weeks to understand. The Compression Lens reframes that gap: the expert is not smarter in the moment — they have better compression. Once you see this, you stop chasing memorisation and start rebuilding the underlying patterns.

Key principles

Compression is lossy — shortcuts hide assumptions, edge cases, and context that experts forgot they know.
Good teaching alternates between compressed insight (the punchline) and decompressed walkthrough (the steps).
Jargon is often a compressed label for a larger idea — define the idea, not just the term.
You know you have understood a topic when you can compress it yourself without losing accuracy.
Different audiences need different compression levels — executives need high compression, implementers need low compression.

How to apply it

1
When you hear a concise explanation, pause and ask: "What experiences had to exist for that sentence to make sense?"
2
Rewrite expert statements as bullet lists of assumptions they did not say aloud.
3
Practice explaining the same idea at three compression levels: one sentence, one paragraph, one worked example.
4
When studying AI concepts, map jargon to concrete behaviours (e.g. "attention" → which words the model weighs when predicting the next token).
5
Use analogies as decompression tools, then test where the analogy breaks — that boundary reveals the real pattern.

Real-world examples

"LLMs predict the next token"

Compressed form of tokenization, probability distributions, training on massive corpora, context windows, and sampling strategies. Decompressing this one sentence is essentially a course in how language models work.

"Use RAG for company knowledge"

Compresses retrieval systems, chunking strategies, embedding models, vector databases, prompt assembly, and freshness/update workflows. The recommendation is simple; the pipeline behind it is not.

Senior engineer's architecture sketch

A five-box diagram on a napkin encodes years of failure modes, scaling lessons, and team constraints. Juniors see boxes; the expert sees incident reports and trade-off decisions.

Common mistakes

Mistaking compression for oversimplification that is actually wrong.
Staying only at the compressed layer and never doing the slow decompression work.
Decompressing forever without re-compressing — you understand pieces but cannot communicate or decide.
Assuming everyone in the room shares the same compression dictionary (especially mixed student/professional audiences).

Key takeaway

Treat every elegant one-liner as a zip file — unzip it until you can rebuild it yourself, then zip it back up for others.