The developers of "Teach My Little Sister How To Drive," an AI-powered driving instruction simulator, are reportedly spending upwards of $1,000 per day to keep the game's demo operational. This significant cost is attributed to the high token usage of the large language model (LLM) powering the game's central chatbot, which has led the studio, Easy Fox, to take out a bank loan.
The game, which places players in the role of a driving instructor coaching a virtual sibling, features an LLM chatbot behind the sibling's character. Players can converse with the AI and issue vocal commands, with the chatbot responding accordingly. The demo, released in February, has seen a surge in players, leading to a more than twentyfold increase in demo participants, according to Easy Fox. This influx has dramatically increased operating expenses.
Easy Fox stated in a Steam post that the daily AI service cost for the demo is over $1,000, not due to individual player expense but the sheer volume of players interacting with the LLM. The studio plans to account for expected AI usage in the final game's price, but has stated players will not be charged separately for token usage after purchase. However, the ongoing unpredictability of AI token costs, which have challenged even large corporations, raises questions about the long-term financial viability.
To mitigate these costs, Easy Fox is exploring options such as implementing local AI models for players with capable GPUs. The developers acknowledge current design challenges include artificial-feeling dialogue and incorrect command execution, issues that could potentially be addressed with traditional voice acting and dialogue systems. The full release of "Teach My Little Sister How To Drive" is planned for January of next year.