Andrej Karpathy, a founding member of OpenAI who joined Anthropic's pretraining team in May, said Thursday on X that he has had success asking language models to explain things in ASD-STE100 — a controlled form of English built to make aircraft maintenance manuals easier to read.

His post walks through four formats for reading model output, moving from controlled writing to diagrams, HTML pages, and explainer videos, and calling each step "even better." As models do more of the legwork, he wrote, "a lot more of our work will rise up the abstractions into oversight and understanding."

The Case For Controlled English

Karpathy wrote that models know ASD-STE100 well and that its "heavy constraints on clean writing style" produce text he often finds "a lot more readable." Because the standard is "quite stringent," he has sometimes softened the request to "80% of the way to ASD-STE100."

ASD, the body representing European aerospace, security, and defense companies, maintains the standard. Its FAQ describes Simplified Technical English (STE) as an international standard for technical writing, built on a set of writing rules and a dictionary of about 900 approved words — most with just one meaning.

A reply under Karpathy's post links to an open-source skill that was already on GitHub and applies STE rules to text that AI agents read. Its README notes the skill isn't built for marketing copy.

Diagrams, HTML Pages, And Explainer Videos

Rather than settling for prose, Karpathy suggests asking for a diagram, which he says can be "a lot easier to process, parse, and understand."

He also recommends requesting output "in HTML" to get an interactive web page. He gave similar advice in May, about a week before he joined Anthropic, when he shared an X post by Anthropic's Thariq Shihipar explaining why members of the Claude Code team use HTML instead of Markdown. A version of that piece ran on the Claude Blog later that month.

The format he is "most bullish on" is custom explainer videos on any topic. His example prompt asks for a "3b1b style" explainer — a reference to the 3Blue1Brown math videos — and instructs the model to use an ElevenLabs API key for the narration. People without a key can ask the model to find free alternatives that run locally. "This is actually starting to work!" he wrote.

Why This Matters

All four suggestions share one mechanism: they change how model output is presented to the person reviewing it. The writing, diagram, and HTML formats work as simple requests to a model; the video example additionally requires an ElevenLabs API key or a free local alternative for narration.

One caveat sits in the standard's own documentation. The ASD FAQ says STE isn't meant for general writing. Still, its underlying principles — short sentences, one topic per sentence, active voice — can carry over to everyday prompts.

Looking Ahead

Karpathy closed the post by urging people to ask models for "large, custom, discardable software artifacts" that "would have never made sense to create before."

The group behind ASD-STE100 distributes the full standard for free, though its FAQ cautions that no tool can replace it. The open-source skill linked in the post's replies states it doesn't include the standard's dictionary — worth checking before assuming any STE wrapper gives complete coverage. For practitioners, the signal to watch is whether format-specific prompting of this kind moves from individual tips into default agent and assistant behavior.