DeepSeek and other Chinese models suffer from translationese; roleplay prompts fix it
dangerous_inference · reddit · 2026-08-18
A Reddit user highlights that Chinese LLMs like DeepSeek often output "translationese"—direct translations of Chinese idioms like "swallowing a whole date" that make no sense in English. Experiments show that while instruction-following varies, enforcing a specific persona (e.g., Data, GLaDOS) eliminates this issue and improves accuracy. The author updated their system prompt to include "NO TRANSLATIONESE" rules and specific examples of common errors.
More from Models
- Prediction: frontier AI revenue growth could slow even as AI usage keeps rising — yi111 · 2026-10-16
- Meta teased 816B-param open-weight Muse Spark Open with 1M context, says meme thread — alexandr_wang · 2026-10-16
- Nolla Health Post-Trains Nemotron 3.5 VL: Visual Diagnosis Jumps from 35% to 66.8% — LoganGrasby · 2026-10-16
- Dev argues coding models now beat nearly all engineers in every branch of software — zack_overflow · 2026-10-16
- Prediction: Chinese frontier models could match GPT-6 by March 2027 — teortaxesTex · 2026-10-16
- Liquid AI's d1-ultra, #1 on Artificial Analysis Decision Index, lands on Vercel — JosephJacks_ · 2026-10-16