This is a rare case when a video game has become a factor in decision-making at a major technology company. Baldur's Gate 3 turned out to be so multi-layered that it confused the algorithms.
The problem with language models is that they often repeat popular errors instead of accurate facts. In the case of a large-scale role-playing game, this leads to failures: multiple patches and hundreds of variables affecting romances and quests confused Grok.
Musk, known for his love of games, didn't want the bot to be wrong about the best classes or the fate of Astarion. Moreover, xAI allegedly created a "war room" to train the AI on League of Legends and bring it to the highest ranks.
The situation became a kind of advertisement for Baldur's Gate 3: if algorithms trained on trillions of data need additional "lessons" to understand the world of Larian, then human creativity is still ahead.