quaint pecan
I am writing to report a significant and persistent bug regarding text rendering on your platform. This issue occurs on both the Kimi AI website and the official mobile application when generating content in Arabic.
1. Detailed Problem Description
The core of the problem lies in the incorrect handling of bidirectional text. When a response is generated in a Right-to-Left (RTL) language like Arabic, and it includes Left-to-Right (LTR) elements such as English words, technical acronyms (e.g., "TPU", "CPU"), brand names (e.g., "Google", "NVIDIA"), or numbers, the text rendering fails.
The result is a jumbled and incoherent output where the word order is broken. Segments of Arabic text are incorrectly placed, and the LTR words disrupt the natural flow of the sentence, making the entire response difficult, and often impossible, to read and understand.
I have attached a screenshot from your website that clearly demonstrates this issue for your reference.
2. Platforms Affected
This rendering bug is consistently present on:
-
The Kimi AI website
-
The Kimi AI mobile application.
3. Steps to Reproduce the Bug
The issue can be easily replicated by following these steps:
-
Enter a prompt that requires a response containing both Arabic and English text. For example, you can use the following prompt:
-
اشرح لي ما هي وحدات TPU التي تطورها شركة Google -
(Translation: "Explain to me what are the TPU units developed by Google").
-
-
Observe the generated output. You will immediately notice the incorrect word order and the jumbled mix of RTL and LTR text segments, just like in the provided screenshot.
4. Impact on User Experience
This bug creates a very poor user experience. The rendering error makes the generated text incomprehensible and extremely difficult to read, which significantly reduces the platform's usability for Arabic-speaking users.
Best regards.
hard scroll
