From IEEE Spectrum:

X.ai released the world’s largest open-source large language model, Grok-1, with 314 billion parameters, surpassing predecessors like Falcon 180B at 180 billion. The timing, post Elon Musk’s suit against OpenAI, raises questions about openness. Grok-1 is under Apache 2.0 license, but lacks documentation and training data, hindering analysis.

Grok-1’s massive size of 314 billion parameters exceeds Meta’s Llama 2 model and could rival GPT-4 and Claude-3 Opus. While size enhances model quality, it complicates deployment and fine-tuning, requiring costly resources. Grok-1 lacks trust and safety features found in fine-tuned models, posing challenges for development and deployment.



Read more at IEEE Spectrum: Grokking X.ai’s Grok—Real Advance or Just Real Troll?