Grok-1 surpasses GPT-3.5 in capabilities, as indicated by the benchmarks provided by the company.
Elon Musk revealed on Saturday the introduction of a new expansive language generative AI model called Grok. Inspired by the “Hitchhiker’s Guide to the Galaxy,” Grok aims to not only respond to a wide array of questions but also, more challenging still, propose the questions themselves!
Grok is set to be integrated into X, formerly known as Twitter, with a unique twist—it’s designed to provide answers laced with humor. In fact, the company encourages users not to engage with Grok if they have an aversion to humor.
With access to real-time knowledge of the world with the help of the X platform, Grok is even capable of answering “spicy questions” that most other AI models reject. This four-month-old generative AI model with 2 months of training is still in the beta phase and the company claims to improve it in the coming days.
According to xAI, Grok has been created to assist humanity to understand and gain knowledge. It is powered by Grok-1 LLM, which has been developed over a period of four months. The prototype Grok-0 was trained with 33 billion parameters, which is said to be as capable as Meta’s LLaMA 2, which supports 70 billion parameters.
Grok, armed with real-time insights from the X platform, possesses the remarkable ability to tackle even the “spicy questions” that many other AI models shy away from. Despite being just four months old and having undergone two months of training, this generative AI model is currently in its beta phase, with promises of continual enhancements in the days ahead.
According to xAI, Grok is a tool designed to aid humanity in understanding and acquiring knowledge. It operates on the Grok-1 LLM, a product of four months of development. The initial version, Grok-0, was trained with an impressive 33 billion parameters, putting it on par with Meta’s LLaMA 2, which boasts 70 billion parameters.
In the realm of benchmarks, Grok-1 (with 33 billion parameters) showcases its prowess, outperforming OpenAI’s GPT-3.5 but falling slightly behind the latest GPT-4. The xAI evaluation further underscores Grok-1’s capabilities as it successfully navigated the 2023 Hungarian national high school finals in mathematics, earning a C grade (59 per cent). Notably, this surpassed the performance of Claude 2 (55 per cent), while GPT-4 secured a B grade with a score of 68 per cent.
These figures highlight Grok-1’s impressive standing, demonstrating that, despite being trained on a comparatively smaller dataset, it can outshine models trained on larger volumes of data and with higher computing requirements. The evolving landscape of AI is indeed fascinating!
Grok-1 underwent training through a tailored training and inference stack grounded in Kubernetes, Rust, and JAX. Its unique feature lies in its internet access, providing real-time information, but with an interesting twist—the company asserts that Grok-1 has the potential to “generate false or contradictory information.” This aspect adds a layer of complexity to its capabilities, reflecting the nuanced nature of its responses.
To address potential challenges in future models, xAI is actively seeking human feedback, a deeper grasp of context, the ability to handle multiple modes of input, and enhanced resilience against adversarial influences. These considerations aim to refine and fortify the capabilities of upcoming models, reflecting a commitment to continuous improvement.
The beta edition of Grok is presently accessible to a select group of users in the United States. Soon, it will extend its availability to X Premium+ subscribers, priced at Rs 1,300 per month, for those who subscribe through a desktop. Stay tuned for broader access and exciting developments!