A Slovenian developer analyzed 2,300 of their own prompts to Claude Code and found that prompting in Slovenian, even with typos and mixed English technical terms, does not significantly degrade performance on larger models. Research indicates that large language models process meaning in a shared, English-centric conceptual space, making the impact of non-English input minimal. While smaller models may show a larger performance gap, the cost in tokens for Slovenian input is also less than expected due to the inclusion of English technical terms and file paths, though output tokens remain a significant factor in latency. AI
IMPACT Suggests prompt language choice has minimal impact on large models, potentially simplifying workflows for non-English speakers.
RANK_REASON Analysis of personal prompt data and existing research on LLM multilingual performance.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
