Part 6/10:
Despite its impressive capabilities, Grock 3 is not without limitations. One of the primary challenges observed was its tendency to "hallucinate" URLs and provide inaccurate citations when using the Deep Search feature. For example, when tasked with generating a report about major AI labs, Grock 3 produced incorrect figures and even overlooked its own company, XAI, in its findings.
This highlights an essential challenge within AI models—ensuring accurate and reliable outputs. Additionally, Grock 3 remains a closed-source model, which constrains public verification of its underlying architecture and training data. It places Grock 3 at a disadvantage compared to open-source models like DeepSeek R1, which allow more transparency and scrutiny.