A parent built shape-based Greek alphabet flashcards for toddlers using GreekLex corpus filtering, ChatGPT candidate triage, and gpt-image-1 illustration generation.
Key Takeaways
Pipeline: downloaded GreekLex (35,304 Modern Greek words), filtered to 3-10 char words with corpus frequency above 100, then batched candidates to ChatGPT to find objects whose shapes echo the letter.
Image generation used gpt-image-1 with the actual Greek letter as a reference attachment; stubborn cases (e.g. phi/snake) required hand-drawn sketches as additional input.
Two card sets produced: object cards (letter-shaped illustration + full word) and alphabet cards (24 letters, reversible by child for memory games).
Kids reportedly learned ~18 letters in one afternoon across two 30-minute sessions; games include a physical “fire game” to reinforce recall.
Eric Carle illustration style was the aesthetic target for all AI-generated card art.
Hacker News Comment Review
The single comment challenges the creator’s claim of being first to make shape-based Greek alphabet cards, citing easy web search results that contradict it – a factual claim the post does not address.
