Elon Musk-Led xAI Unveils Grok, Here's Everything You Need To Know

Elon Musk-led startup xAI Corp. has announced a new generative AI model dubbed Grok. To recap, the 52-year-old tech mogul formed the xAI artificial intelligence company earlier this year "to unlock mysteries of the universe".

Notably, Grok has been announced ahead of OpenAI's first developer conference, which is slated to take place in San Francisco on November 6.

Grok is expected to compete with existing LLMs, including ChatGPT and Google's LaMDA. Interestingly, Grok has been developed by ex-engineers from OpenAI, Deepmind and Google.

Musk says Grok has a sense of humour and is maximally and truly curious. Here's everything we know about Grok so far:

Grok: Availability, subscription fee

With just 2 months of training, Grok is currently in the closed beta stage. However, the AI tool will be available to select US-based X users with an invite.

US-based users who are verified on X can sign up for the waitlist on the xAI website. According to Musk, Grok will be available to X Premium+ subscribers in the coming days. To those unaware, X Premium costs £8 a month and £84 a year. When subscribed from a desktop, the Premium+ tier costs £16 a month and £168 a year in the UK.

Everything else we know about Grok so far

xAI announced Grok on Musk-owned social media platform X, noting that the generative AI product is "modelled after the Hitchhiker's Guide to the Galaxy". In other words, Grok is designed to answer almost everything with a sense of humour, coupled with a rebellious streak.

Furthermore, the AI company confirmed that Grok will be able to access X's (formerly Twitter) real-time data. It is worth noting that Musk previously confirmed that xAI will use Twitter and Tesla data to train AI systems.

The generative AI model introduced by xAI will not shy away from answering questions that are often rejected by other AI bots like ChatGPT and Google Bard.

If the screenshots of early benchmark data shared by xAI are anything to go by, Grok-1 has already surpassed GPT-3.5 and Inflection-1. Notably, Grok-1 is the LLM behind Grok.

On the downside, Grok failed to outperform Google's Palm-2 and GPT-4. However, xAI clarified that they were able to train Grok only for 2 months. So, it is safe to assume that the AI model will get better over the next few weeks.

Moreover, Grok underwent a slew of evaluations to measure its math and reasoning abilities. Unsurprisingly, the AI model was able to perform most of the other models in its compute class.

GPT-4, which boasts years of training data and computing resources was able to beat Grok hands down. Aside from this, xAI put Grok through the 2023 Hungarian national high school finals in mathematics test.

The generative AI model got 59 per cent (C grade). For comparison, GPT-4 got a B grade, scoring 68 per cent and Claude-2 achieved a C grade with 55 per cent in the above-mentioned test.