Updates from the development team for Teach My Little Sister How to Drive announce new content and explain ongoing technical challenges. Recent changes include facial emotional expressions and more hair customization, plus tutorial fixes so progression no longer stalls. Developers acknowledge player complaints about overly long, unnatural replies and incorrect in-game commands; these stem from heavy reliance on a fallback OpenAI model after their intended primary model, Google’s Gemini, hit usage limits. The team is prioritizing restoring reliable Gemini access, improving conversational quality of the current model, and adding local AI options for players with capable GPUs. They also request specific dialogue feedback on Discord to make the little sister feel more natural across languages.
Rapid, unexpected demo growth - more than twentyfold - has sharply raised AI costs (reported over $1,000 per day), prompting a bank loan to keep the demo running and a plan to include expected AI usage in the paid release price rather than charging per-player. Multiple patches (1.2.4-1.2.9) addressed issues such as excessive dialogue length, AMD/Ryzen crash fixes, a tutorial seatbelt/engine bug, region-specific model routing for Mainland China and Russia, and swapping fallback models for stability. The team is collecting crash logs and testing alternative models for smoother regional play, while warning the demo could close early if financial pressure continues.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.