Why can text models draw letters? ASCII art in training data may be the answer
giffmana · x · 2026-10-27
giffmana argues that models capable of drawing an "A" likely saw training samples describing how letters are constructed—two oblique lines touching on top with a horizontal bar at mid-height. He bets on ASCII art in the training data as the cause. His experiment: same drawing prompt across models, 5 attempts each picking the best, excluding vision-trained models (hence no Gemma 4). He admits it's just a guess.
Related event: Text-Only LLMs Tested at Drawing Pixel Art, Qwen3-32B Leads(4 posts)→
More from Models
- DeepSeek Harness plugin ecosystem is booming as cheap, mid-tier-but-solid model draws tinkerers — op7418 · 2026-10-27
- Uncensored Wan 1.3B fine-tune hits Hugging Face, runs video gen on 8GB VRAM — Promptmethus · 2026-10-27
- The tokenizer tax: non-English AI users face up to 17x token cost gaps — alexvoica · 2026-10-27
- Blogger calls deepseek-v4.1-flash-1017 the most underrated open-source model of the year — intellectronica · 2026-10-27
- Photographer finds ChatGPT now refuses watermark removal after earlier doing it easily — moisesg88 · 2026-10-27
- Gemini 4 Flash surfaces internally, codenamed 'Hydrogen', expected soon — koltregaskes · 2026-10-27