炊紙(kashikishi) is a text editor that utilizes GPU to edit text in a 3D space

Original Article ↗ Hacker News Discussion ↗

A GPU-accelerated text editor for Japanese, Kashikishi, showcases experimental 3D and AR-style rendering of text, including proper vertical layout, prompting interest in how non-Latin scripts and spatial interfaces could reshape editing workflows. Commenters weigh its playful, animated “physical” interactions against the needs of power users who prioritize speed and minimalism, and branch into related issues such as machine-translation quality (Google Translate vs GPT/DeepL), the language barrier around Japanese open source projects, and the aesthetics of different writing systems.

Project & Demo Impressions

  • Kashikishi is a GPU-accelerated text editor that can render text in 3D, including psychedelic and AR modes (shown in GIFs from the repo).
  • Some viewers of the demo video mostly saw standard 2D text and wished for clearer 3D examples.
  • The psychedelic mode is specifically highlighted as visually striking, especially for Japanese glyphs.
  • There’s curiosity about AR/VR integration; some note parallel experiments that render code or text in 3D/AR.

3D / AR UI Potential vs Practicality

  • Enthusiasts see 3D and mixed reality as a chance to completely rethink UI, beyond flat windows and tabs.
  • Ideas mentioned: zoomable code spaces, physical-feeling animations, tangible tools (e.g., a virtual timer you wind, record-like album selection), and 3D visualization of code structure.
  • Others prefer traditional, instantaneous text editing without animations, valuing speed and simplicity for serious writing.

Japanese Text, Vertical Writing, and Scripts

  • A key motivation: good vertical-writing support for Japanese, which many editors lack; instant switching between vertical and horizontal is seen as valuable.
  • Several comments appreciate Japanese script aesthetics and compare them to Mongolian, Tibetan, Arabic, and historical Latin scripts.
  • Some argue that technology has flattened and simplified scripts visually; others counter that simplification improves legibility and access.

Language, Translation, and Tooling

  • The README being in Japanese prompts discussion of machine translation quality (Google Translate, GPT-4o, DeepL, Gemini).
  • Issues reported: inconsistent handling of proper names, inserting/removing “not,” hallucinated content, and grammar-driven subject insertion in some language pairs.
  • Some prefer slightly worse but “honest” translators over fluent but hallucination-prone ones; others note humans also “hallucinate” in translation.

Open Source & Community Aspects

  • Maintaining a project in a non-English language is seen as a way to reduce spammy PRs, but also risks fewer useful contributions.
  • Examples given of widespread low-value PRs (e.g., trivial README edits, contest-driven spam) on popular repos.