I'm a self-taught data scientist, which is a polite way of saying I learned the math from YouTube and the English from context clues, and the second one didn't fully take. My technical skills are fine. My academic-reading English is not. Opening a machine-learning paper felt like swimming in cold deep water fully clothed: I'd understand the surface, hit a phrase like 'non-parametric approximation' or 'latent representation,' sink for a second, claw back up, and then keep reading without fully understanding the sentence I'd just survived. After three sentences of that I'd close the tab and tell myself I'd read it 'later,' a word that in this context meant never.
The friction was the vocabulary, not the ideas
Here's what took me embarrassingly long to figure out: I wasn't failing to understand the concepts. I was failing to understand the words the concepts were wearing. The ideas in a lot of these papers are things I already half-knew from building models. But every paragraph had two or three terms that stopped me cold, and stopping cold every paragraph turns a forty-minute read into a two-hour ordeal that I'd abandon out of sheer fatigue. The friction wasn't intellectual. It was vocabulary, and vocabulary is a solvable problem.
So I flipped the order. Instead of looking words up mid-sentence and losing the thread each time, I now process the vocabulary before I read a single line of the actual argument. I paste the abstract and the first section into Gemini 3 Pro Preview — which handles long passages without losing the plot — and run the Build a Vocabulary Drill with Definitions, Examples, and Answer Key prompt. It pulls the ten hardest domain-specific terms, defines them plainly, shows each in a second context, and flags the confusing ones. I spend eight minutes on the drill. Then I read the paper like a person.
The single best column in the output is 'In the passage,' where it quotes the actual sentence the word appeared in. A dictionary gives you a definition in a vacuum; this gives me the word doing its job in the exact text I'm about to read, so when I hit that sentence for real I already know what's happening. It's the difference between studying a vocabulary list and studying the specific words that are about to block me. Generic word lists never worked for me precisely because they weren't tied to anything — these are tied to the paper sitting in my other tab, which is the whole reason they stick.
Why the fill-in-the-blank matters more than the definitions
I assumed the definitions would be the valuable part. They're useful, but the part that actually made the words stick is the fill-in-the-blank exercise. Reading a definition is recognition — you nod, you feel like you learned it, you forget it by the next paragraph. Filling a word into a new sentence is recall, and recall is where memory actually forms. By the time I've done eight fill-in-the-blanks, the words aren't 'things I just looked up,' they're words I've used, and using a word once is worth reading its definition five times.
- Reading with the vocabulary pre-processed is roughly three times faster than stopping to look things up mid-sentence
- The fill-in-the-blank forces recall, not recognition — that's the difference between feeling like you learned it and learning it
- The 'watch out for' column saved me real confusion: 'confound' the verb and 'confounding variable' the noun are not the same idea
- Doing the drill on the abstract first means I know the paper's key terms before I hit them in the dense middle sections
The result, and the honest limit
I read six papers this month. For comparison, last month I read zero, because the vocabulary friction made a thing I enjoy feel like unpaid homework. The prompt did not make me smarter and it did not teach me machine learning — I already knew the machine learning. It removed the specific, stupid, solvable obstacle that was sitting between me and something I actually want to do, and removing friction from a thing you like is how a chore quietly turns back into a hobby.
The honest limit: for the most technical terms, the plain-language definition is sometimes a simplification I have to firm up later by reading the paper itself, which is fine, because by then I can actually read the paper. I treat the drill as a launchpad, not the final word — it gets me past the wall so the paper can teach me the precise version. I've also learned to set the focus type deliberately: 'domain-specific terms' for a dense methods paper, 'high-frequency words I likely don't know' for older or more rhetorical writing where the hard words are general English, not jargon. Matching the focus to the kind of text is the one knob that most changes the quality of the drill.
When a paper introduces a genuinely new concept rather than just a new word for a familiar one, I'll follow up with my Explain Any Concept at Two Levels (Kid Version + The Honest Version) prompt to build the intuition before I wade into the dense version. Vocabulary first, intuition second, then the actual paper — that order turned reading from a wall into a ramp. Grab the Build a Vocabulary Drill with Definitions, Examples, and Answer Key prompt on Prompt Dock, point it at whatever you've been pretending you'll read 'later,' and read it now instead.