Elon Musk's xAI Reveals Grok, An AI Chatbot That Answers 'Spicy Questions'

Elon Musk-backed xAI has finally released its first product, an AI chatbot that is claimed to be more humorous than ChatGPT. The bot is named Grok, a term invented by writer Robert A. Heinlein as a meaning-fluid Martian word, though xAI says it is inspired by "Hitchhiker's Guide to the Galaxy." The company claims Grok AI is "intended to answer almost anything" and will even go as far as telling users what to ask.

Grok is a product of merely weeks' worth of development and is still in a "very early beta phase," according to the company. This explains why xAI is keeping it exclusive to a select group of individuals who are based in the U.S. and also happen to pay for the X Premium+ subscription priced at $16 per month. For now, the standout Grok feature appears to be its humorous and witty tone of conversation, which is something that has been baked in at its core. 

"It will also answer spicy questions that are rejected by most other AI systems," xAI writes in its blog post. That's a double-edged sword and could backfire, especially when the company has spent merely two months training it. The large language model powering Grok is called Grok-1 and it was trained to handle 33 billion parameters, nearly half that of Meta's own Llama 2 model.

The road ahead for Grok

Musk says that Grok will be available as an exclusive feature on X, formerly Twitter, and will also be available as a standalone app. xAI claims Grok outperformed Llama 2 and OpenAI's GPT 3.5 at tests such as the GSM8k comprising middle-school mathematics problems, multidisciplinary multiple choice questions (MMLU), and Python coding-oriented tasks in HumanEval.

However, Grok-1 is still a healthy margin behind seasoned players like OpenAI's GPT-4 powering the latest iteration of ChatGPT, Google's PaLM 2, and Anthropic's Claude 2 models. The biggest difference happens to be the size of the training dataset and the amount of time poured into refining the project, both machine and human-assisted kind.

xAI is working to solve that problem by hiring human experts from different domains to improve it. Grok has one key aspect in its favor, though: the chatbot is connected to the internet and pulls data in real time from X. In contrast, OpenAI waited months before it connected ChatGPT to the internet through web plug-ins. But once again, it's a risky endeavor, as false information that often goes viral could very well dictate Grok's response until it's flagged and corrected. "We will work towards developing reliable safeguards against catastrophic forms of malicious use," assures xAI. Right now, Grok only supports text-based interactions, but the company says it has plans to add multi-modal capabilities down the road so that the AI chatbot can also process image and audio inputs, as well.