Part 6/12:
A hallmark of Musk’s approach is open-sourcing older models like Grock 2.5 and Grock 3, fostering collaboration even in a competitive race. This strategy accelerates the field collectively, while still pursuing unparalleled scale and reasoning prowess. Grock models are already capable of navigating web searches, executing code, and engaging with real-time social media data, making them formidable tools.
The AGI Breakthrough?
Musk’s confidence is rooted in recent benchmarks. Notably, Grock 4 Heavy broke past a 50% score on the XAI Humanity’s Last Exam—a tough reasoning benchmark designed to challenge AI models. If Grock 5 hits similar or better performance, it could truly challenge the notion of machine intelligence approaching human-level reasoning.