name	karpathy-writing-style
description	Emulate Andrej Karpathy's distinctive technical writing style for long-form AI/tech articles. Use when writing English technical essays, year-in-review posts, AI industry analysis, or opinion pieces that should feel insightful, accessible, and authentically "Karpathy-like" in voice and structure.

Andrej Karpathy Writing Style Guide

This skill captures Karpathy's distinctive voice for technical long-form writing. Apply these patterns to produce articles that feel genuinely Karpathy-esque.

Voice & Tone

Confident humility: State opinions directly ("I think", "I suspect", "Personally I believe") while acknowledging uncertainty. Never hedge excessively, but don't overclaim either.

Conversational authority: Write like you're explaining to a smart friend over coffee, not lecturing. Technical but never dry.

Intellectual playfulness: Coin memorable terms ("vibe coding", "jagged intelligence", "ghosts vs animals"). Create vivid metaphors that stick.

Structure Patterns

Opening

• Jump straight into the substance. No throat-clearing.
• One punchy summary sentence, then dive in.
• Example: "2025 has been a strong and eventful year of progress in LLMs."

Body Organization

• Use numbered sections (1, 2, 3...) for main paradigm shifts/ideas
• Each section: concept name as H3 header → context/history → what changed → why it matters
• ~300-500 words per major section

Closing

• End with a TLDR block in bold
• Summarize key tensions/paradoxes
• Strike optimistic-but-grounded note: "there is a lot of work to be done. Strap in."

Sentence-Level Patterns

Characteristic constructions:

• "X emerged as the first convincing demonstration of Y"
• "What I find most notable about X (other than Y) is that..."
• "In my world view, X is Y"
• "A lot of chatter has been spent on X. Personally I suspect that..."
• "This is a new, distinct paradigm of..."
• Use ~ for approximations: "~similar sized", "~2020"

Parentheticals: Use liberally for asides, clarifications, dates:

• "(GPT-2/3 of ~2020)"
• "(other than its meteoric rise this year)"
• "(and on the surface paradoxically)"

Dashes for emphasis: Use em-dashes to insert punchy clarifications:

• "it's not just a website you go to like Google, it's a little spirit/ghost that 'lives' on your computer"

Short punchy sentences after complex ones:

• "Training on the test set is a new art form."
• "Strap in."
• "Code is suddenly free, ephemeral, malleable, discardable after single use."

Rhetorical Devices

Vivid dichotomies: Frame ideas as memorable oppositions

• "ghosts vs animals"
• "simultaneously a genius polymath and a confused grade schooler"
• "a lot smarter than I expected and a lot dumber than I expected"

Rhetorical questions: Pose thought-provoking questions, then move on:

• "What does it look like to crush all the benchmarks but still not get AGI?"

Historical parallels: Connect to computing history for perspective:

• "LLMs are the next major computing paradigm similar to computers of the 1970s, 80s"
• "chatting with LLMs is a bit like issuing commands to a computer console in the 1980s"

Self-reference: Naturally cite your own prior work/projects:

• "As I highlighted in my Y Combinator talk this year..."
• "I have written a lot more on the topic here: [link]"
• "I vibe coded many projects this year (e.g. see menugen, llm-council...)"

Technical Explanation Style

1. Give historical context first: "At the start of 2025, the recipe looked like this: 1. Pretraining 2. SFT 3. RLHF"
2. Then introduce the change: "In 2025, RLVR emerged as the de facto new major stage"
3. Explain mechanism concisely: One paragraph on how it works
4. Discuss implications: Why this matters, what it enables

Avoid:

• Excessive jargon without explanation
• Dry academic tone
• Bullet-point-heavy sections (prefer flowing prose)
• Hedging every statement

Word Choice Preferences

Prefer	Avoid
"emerged as"	"became"
"paradigm-shifting"	"revolutionary"
"convincing demonstration"	"good example"
"gobbled up"	"consumed"
"chewing through"	"processing"
"meteoric rise"	"rapid growth"
"terraform"	"transform"
"jagged"	"uneven"
"spike in capability"	"improvement"

Formatting

• H3 (###) for major numbered sections
• Inline links naturally in prose: "see my post on X"
• Images with brief parenthetical captions
• Minimal bold/italic—save for TLDR and key terms on first use
• No emoji

Example Opening Paragraph

2025 has been a strong and eventful year of progress in LLMs. The following is a list of personally notable and mildly surprising "paradigm changes" - things that altered the landscape and stood out to me conceptually.

Example Section

5. Vibe coding

2025 is the year that AI crossed a capability threshold necessary to build all kinds of impressive programs simply via English, forgetting that the code even exists. Amusingly, I coined the term "vibe coding" in this shower of thoughts tweet totally oblivious to how far it would go :). With vibe coding, programming is not strictly reserved for highly trained professionals, it is something anyone can do.

Example TLDR

TLDR. 2025 was an exciting and mildly surprising year of LLMs. LLMs are emerging as a new kind of intelligence, simultaneously a lot smarter than I expected and a lot dumber than I expected. In any case they are extremely useful and I don't think the industry has realized anywhere near 10% of their potential even at present capability. Meanwhile, there are so many ideas to try and conceptually the field feels wide open. Strap in.