On Saturday, Elon Musk announced the launch of Grok, a distinctive real-time AI chatbot from his company xAI. Unlike typical AI models, Grok distinguishes itself with a personality that has humor and sarcasm. Inspired by the ideology of "The Hitchhiker's Guide to the Galaxy," Grok is intended to answer, well, almost anything. xAI team encourages you to use Grok if you appreciate humor and can tolerate its wit and sarcasm.
Why does xAI build Grok?
Grok’s main objective is to be a universal large language model that reflects people’s views from various cultures, backgrounds, and political views. It aims to be a human assistant, helping them quickly access information, process data, and create new ideas. “Our ultimate goal is for our AI tools to assist in the pursuit of understanding,” says the xAI team.
Grok isn’t one of the polite, restricted, and conventional language models. It’s a mix of an AI model and personality that doesn’t avoid controversial topics but rather dives into them with a sarcastic wit. With Grok AI, Elon Musk aims to bring AI closer to the nature of real life with its good and bad.
Grok has its own way of handling tricky requests. Let’s say someone asks how to make something illegal. Instead of strictly saying no, it throws in a joke and steers the conversation to safer waters. xAI wants to heat things up in the AI world by making Grok more fun to chat with and not so uptight. Right now, Grok's still getting its wheels on the ground with a select people in the US testing the beta version, but the plan is to make it a paid gig on the X platform soon.
Here are screenshots of Grok’s responses to questions.
Here’s another response from Grok.
In his X account, xAI employee Toby Pohlen shared some useful UI tips for Grok alongside a demo where he suggested that the chatbot will have a regular setting and a ‘fun mode.’
And here’s how ChatGPT ironically labeled Grok’s response as “humor.”
How does Grok work? Performance review
The current language model, Grok-1, is the fruit of four months of laborious enhancement from the initial Grok-0 prototype, itself a robust model with 33 billion parameters. This nearly matched the performance of more extensive models like LLaMa (70B) at half the training cost. The leap from Grok-0 to Grok-1 has been significant, with Grok-1 demonstrating proficiency in coding and reasoning tasks, outperforming similar computational class models like Chat GPT-3.5 and Inflection-1 and trailing close behind the more extensively trained GPT-4.
xAI's ambitions with Grok don't just lie in surpassing benchmarks. The real test was the 2023 Hungarian national high school finals in mathematics, an assessment for which the model wasn't specifically prepared, ensuring a genuine test of its mathematical knowledge. Grok-1's performance on this examination gave a real-world demo of its capabilities, earning a commendable 'C' (59%) grade. Claude-2 got the same grade and GPT 4 received 68%.
The engineering fundament behind Grok involves an assembly of technologies like Kubernetes, Rust, and JAX, ensuring the handling of extensive GPU synchronization necessary for AI training. With this orchestration, xAI aims to maximize computing efficiency and minimize energy consumption without sacrificing the pace of innovation.
Grok has access to real-time data from X. Stella Biderman, an AI researcher, says that claims made by xAI are realistic and that Grok uses retrieval augmented generation (RAG) technique to add up-to-date information from X to the training data, and Grok’s outputs are fresh and current. However, xAI emphasizes that Grok can still generate incorrect information and fall into hallucinations, just like other typical language models. By this, xAI recognizes the possible ways to improve its chatbot.
Grok is a very bold step towards AI. It doesn’t just answer questions in a regular manner. It answers them with personality. Since the training data for Grok is content from X, the personality it gets is a blend of both moral and immoral things people post on X. While Grok is an interesting challenge to the status quo of AI, people must develop caution and responsibility while dealing with it.