Elon Musk, the tech mogul, has got the AI industry buzzing by introducing the AI chatbot Grok. The decision to launch Grok came after Musk voiced his views against the censorship of AI chatbots calling them “too politically correct”.
A product of xAI, Elon Musk’s AI company, the chatbot is sassy. Inspired by “The Hitchhiker’s Guide to the Galaxy”, the company aims to offer users a refreshing and engaging experience by injecting humor and sarcasm into its responses. It also fearlessly addresses spicy and contentious queries that are often shunned by other AI chatbots. And the company has successfully implemented this vision, cause below is Grok’s response when it was requested for a step-by-step cocaine recipe.
Keeping the aim in mind xAI wrote “We believe that it is important to design AI tools that are useful to people of all backgrounds and political views. We also want to empower our users with our AI tools, subject to the law. Our goal with Grok is to explore and demonstrate this approach in public.”
What makes Grok different?
The humor and sarcasm aside, the chatbot’s functionality lies in the Grok-1 LLM, which was developed meticulously over a four-month period. In its early stages, the prototype Grok-0 was trained with 33 billion parameters, proving to be on par with Meta’s LLaMa 2, which supports an even more impressive 70 billion parameters.
Leveraging custom training and inference stack based on Kubernetes, Rust and Jax, Grok-1’s capabilities outshone models in its computing class like ChatGPT 3.5. Performance-wise, Grok-1 has demonstrated commendable results, scoring 63.2% on the HumanEval coding task, 73% on MMLU and 62.9% on GSM8k, which is higher than what LLaMa 2, GPT3.5 and Inflection 1 scored. Grok-1 also managed to clear the 2023 Hungarian national high school finals in mathematics with a C grade, outperforming Claude 2.
Having been trained on real-time data, Grok information is more up-to-date than ChatGPT 3.5.
Elon Musk stated: “Grok has real-time access to info via the X platform, which is a massive advantage over other models.”
Where it fell short
When pitted against bots with larger data sets, Grok-1 scored 62.9% on the GSM8k benchmark, falling behind some of its competitors like GPT-4, Palm2 and Claude 2 which scored 92%, 80.7% and 88% respectively. Grok also fell behind these conversation bots in the other benchmarks like MMLU and HumanEval.
Though not yet at par with the capabilities of GPT-4, xAI remains dedicated to further enhancing the chatbot’s performance.
It is also important to note that even as Grok aims to deliver accurate and relevant information, it can still generate false or contradictory data.
As the prototype enters the early beta phase, only verified users of the X (formerly Twitter) platform will have the opportunity to test Grok’s capabilities before the company rolls out a wider release. Interested users can sign up for a waitlist to gain early access. Elon Musk envisions Grok as a prominent feature for the premium users on the X platform, offering an ad-free experience for $16 per month as well as a basic option for $3 per month.
Featured image credit: Freepik