A parent built visual-association flashcards for kids learning Greek, where each object is drawn to resemble the letter its name starts with.
Key Takeaways
Pipeline: GreekLex corpus (35,304 words) filtered by length (3-10 chars) and frequency (100+), then batched into ChatGPT for shape-match candidates, then rendered with gpt-image-1 (gpt-image-1.5).
Frequency columns in GreekLex let the author target vocabulary kids would plausibly recognize, cutting 50-2,500 candidates per letter to 10-200.
Stubborn cases like phi (snake shape) required hand-drawing first, then asking the image model to restyle it, bypassing prompt-only failures.
Two card sets: object cards (letter + illustration + word) and 24 alphabet cards; games include a memory match and a physical “fire game” to reinforce recall.
Author claims these are the first shape-echo Greek alphabet cards; English equivalents exist (Letterland) but rarely use true shape-mimicry.
Hacker News Comment Review
Commenters with native Cyrillic backgrounds noted partial overlap with Greek letters eased learning, but ξ remained universally hard to draw reliably.
Several commenters suggested this method scales to other scripts; Chineasy was cited as a parallel approach for Hanzi.
Practical debate emerged over whether single-letter cards are sufficient, with one commenter arguing syllable-grouped teaching is needed for real reading fluency.
Notable Comments
@gschizas: Provides a full per-letter rating of the card choices, flagging weak matches and crowdsourcing alternatives from Greek speakers.
@jaharios: Suggests χελιδόνι (swallow) for chi, with X-shaped tail-wings as a stronger visual match than the card’s choice.
