Did the restricted-English approach preserve all the information?

In the author's Japanese test, 99 listed facts became 95, with the four losses limited to things like file names. Average sentence length fell from 41 to 28 characters, but total text grew 14%. A separate English experiment by developer Lucian Ghinda in August found Claude lost 46.8% of code-specific facts when told to follow ASD-STE100.

How was the video made in the test?

The author asked Claude Code for a 60-second 3Blue1Brown-style animation as a Manim Python script with narration for each scene. Six narrated scenes came out; rendering elsewhere produced a 60-second video. For something easier, Gemini Notebook can generate a video by loading the material, while Manim takes more setup.

Large Language ModelsITmedia AI＋10m agoOct 14, 2026, 06:01 JST

Karpathy: four ways to cut the time spent reading AI output

Send on LINE

3 Key Points

What happened
Karpathy posted on X on October 2, 2026 that understanding LLM output will take far more of our time, and ranked four fixes — prose, diagrams, web pages, video. The post drew over 7.5 million views.
Why it matters
The bottleneck is shifting from writing to understanding, so the request format itself, not the reader's effort, becomes the way to cut the load. The author tested all four and found the video, built last, was unexpectedly good.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

FAQ

Did the restricted-English approach preserve all the information?: In the author's Japanese test, 99 listed facts became 95, with the four losses limited to things like file names. Average sentence length fell from 41 to 28 characters, but total text grew 14%. A separate English experiment by developer Lucian Ghinda in August found Claude lost 46.8% of code-specific facts when told to follow ASD-STE100.
How was the video made in the test?: The author asked Claude Code for a 60-second 3Blue1Brown-style animation as a Manim Python script with narration for each scene. Six narrated scenes came out; rendering elsewhere produced a 60-second video. For something easier, Gemini Notebook can generate a video by loading the material, while Manim takes more setup.

ITmedia AI＋Read Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.