AI companies often have distinct areas of focus. For example, OpenAI has historically catered to general consumers, whereas its competitor Anthropic primarily serves business clients. Recently, it came to light that Elon Musk’s xAI has been dedicating significant attention to video-game guides. A comprehensive report published on Friday delved into xAI, the AI firm recently acquired by SpaceX, highlighting how Musk’s management style has created challenges for staff. One notable story from the report involved a model launch being postponed last year.

The delay occurred because Musk was unhappy with how the chatbot handled detailed inquiries about the game “Baldur’s Gate,” according to sources familiar with the situation. Senior engineers were reassigned from other tasks to refine the responses before the release. It’s easy to sympathize with a skilled engineer expecting to work on advanced AI challenges, only to be diverted into assisting a 54-year-old with his gaming hobby. This incident prompts a further question: Did Musk ultimately achieve the gaming assistance he sought?

To explore this, an RPG enthusiast compiled five general questions about Baldur’s Gate and tested them on xAI alongside three leading models in an informal evaluation dubbed BaldurBench. For full transparency, all chat transcripts are publicly available: Grok, ChatGPT, Claude, and Gemini. On a positive note, Grok provided quite helpful information. Its answers were packed with gaming terms like “save-scumming” for saving and “DPS” for damage, but they were knowledgeable and practical for those familiar with the jargon.

Grok also showed a strong preference for tables and theorycraft, which is typical for game guides. Since many Baldur’s Gate guides exist online, the models generally pulled from similar sources, resulting in mostly stylistic differences. ChatGPT favored bulleted lists and concise phrases, while Gemini emphasized key terms in bold. Claude offered the biggest surprise by being cautious about spoilers. When asked about effective party compositions, it concluded by advising, “don’t stress too much and just play what sounds fun to you.”

It’s worth remembering that, based on earlier reporting, xAI has specifically aimed to match competitors in this domain. Therefore, the fact that Grok’s guidance ended up similar to other models after the reported push shouldn’t be overinterpreted. Still, it’s reassuring to see that xAI can deliver results when it concentrates its efforts.

EMBED_PLACEHOLDER_1