glyph8192: 8,192 alien glyphs with a reversible custom tokenizer, open-sourced
beffjezos · x · 2026-10-11
Inspired by the alien alphabet in Grimes' new Accelerando MV, the author built glyph8192: 8,192 original glyphs paired with a reversible custom byte-pair tokenizer, letting machines "write" directly in token space instead of Latin letters.
Key details: token ID equals glyph ID across the whole vocabulary; the tokenizer derives from the first 8,192 mergeable ranks of o200kbase, preserving its pretokenization pattern and all 256 byte values (not the full OpenAI tokenizer); no embeddings or model inference required. The project ships a searchable character archive, a local browser encoder, downloadable fonts/atlases, and an encoded-lyrics specimen.
More from Fun
- AI digs through 400 years of Dutch East India archives, finds new dodo eyewitness account — _jaydeepkarale · 2026-10-11
- Musk shows Grok researching and ordering Lego Star Wars kits in one prompt — elonmusk · 2026-10-11
- Suno 4.5 Song Meets Claude-Made Slopcore Video, Surprisingly Polished — technollama · 2026-10-11
- CoLM's 'Most Fun Paper' Gets a Nod From Researchers on X — YouJiacheng · 2026-10-11
- Multiple Google-penalized sites show recovery signs after late-September ranking shift — gaganghotra_ · 2026-10-11
- Musk touts Grok Bot as Nicki Minaj says it 'saved my life' and builds company assistants — elonmusk · 2026-10-11