Is it worth writing to AI agents in English to save tokens?
The author investigates the common advice that using English when interacting with AI agents saves significantly more tokens than using Russian. After conducting 40 runs across five routine tasks, the author compared efficiency and resource consumption. The results showed that while Russian does require more tokens, the difference averages only 12 tokens per task, which is negligible (about 0.005% of the total data). Task performance was identical for both languages, with 17 out of 20 successful completions each. The author notes that the variance between identical runs can reach up to 3.4 times, which is far more significant for a budget than language choice. The article provides a detailed breakdown of where language-based savings might be justified and where they are practically meaningless.
This is a summary. Read the full article at the original source:
HabrRelated stories
A well-formed number is not a measurement: 13 defects across 7 eval tools
A recent study highlights a critical failure mode in evaluation software for LLMs and AI agents: the tendency to produce well-formed numbers that lack…
A growing movement of anti-AI activists is increasingly turning to direct action to protest the rapid development of artificial intelligence. Groups l…
Google may expand Gemini 'Call for Me' to personal contacts
Google is reportedly planning to expand its AI-powered 'Call for Me' feature, which currently assists users with business-related phone calls, to incl…


