SB Intuitions releases Sarashina free beta; translation test shows natural Japanese but meaning shifts
SB Intuitions, a SoftBank subsidiary, launched a free public beta of its Japanese-focused generative AI Sarashina on October 13, 2026. The service is accessible via browser with a Google account at chat.sarashina.ai.
Key points
- SB Intuitions launched Sarashina free beta on Oct 13, 2026, accessible via Google login
- Sarashina's Japanese translation felt most natural but reversed logic in a key sentence
- Claude and Gemini kept closer to English structure, possibly prioritizing accuracy for research
Sarashina produced the most natural-sounding Japanese, using phrasing like "honko" (this paper) and "urazuiteiru to ketsuronzuketeiru" (concludes that it supports). However, careful comparison with the English original revealed three meaning shifts in Sarashina's output: it rendered a causal relationship as concessive ("despite having no predictive power" instead of "because"), translated a theoretical "should" as a normative "must not," and used "rational approximation" (gōriteki) where "appropriate approximation" (datō na kinji) avoids confusion with the economics term "rational expectations." Claude and Gemini kept closer to the English structure, which the author suggests may reflect intentional faithfulness for research papers.
The author notes Sarashina's chat has character limits requiring text splitting for full papers, and the beta terms state user data may be used for model improvement. The comparison covers one abstract with one translation per model; results may vary by text and field.
Comparing translations of a Nobel Prize in Economics paper using Claude, Gemini, and the domestic AI 'Sarashina'. Naturalness and accuracy are different things.
note.com · 14 October 2026
Loading the full article…
This text was published by note.com and written by K_AI. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Meta introduces Muse Wonder and Muse Wonder Full frontier models for agents · 29 src
- Google launches Gemini 4 Argon model globally for Pro and Ultra subscribers with introductory API pricing · 29 src
- TypeSafe AI releases Jev 1.14 with images, box type and 262k token context · 1 src
- The Nordics and NVIDIA Show What’s Possible When a Region Commits to Advanced Computing · 1 src
- LFM2.5-24B-A2B: Built for fast on-device agents · 1 src
Comments
via GitHub Discussions