Grok AI: A humorous challenger to ChatGPT- know its capabilities and more

Elon Musk's Grok AI, designed with humour and knowledge in mind, is here to answer your questions. Discover its capabilities and availability in this brief overview.

Meet Grok- the AI that adds humour and knowledge to your questions. (via REUTERS)

Elon Musk has introduced Grok, a unique AI model, designed to reply to prompts with a dash of humor. Grok is integrated with the X platform. If you are not a fan of humour, this AI may not be your cup of tea, says xAI. Grok leverages real-time knowledge from the X platform, empowering it to tackle even the spiciest of queries. Although it's still in beta, this AI model, with two months of training under its belt, aims to refine and expand its capabilities over time.

Purpose of Grok AI

The goal behind Grok AI, as per xAI, is to help people understand and learn. This AI is powered by Grok-1 LLM, developed over four months. Its prototype, Grok-0, was trained with 33 billion parameters, putting it below Meta's LLaMA 2, which uses 70 billion parameters.

We are now on WhatsApp. Click to join.

Grok's Abilities

Grok-1's capabilities are impressive. It scores 63.2% on the HumanEval coding task and 73% on MMLU. While it may not outshine GPT-4, xAI claims that Grok-1 has improved significantly compared to Grok-0 in a short time.

On the GSM8k benchmark, which focuses on middle-class maths word problems, Grok-1 achieves 62.9%, surpassing GPT-3.5 and LLaMa 2 but falling behind Palm 2, Claude 2, and GPT-4.

Grok-1 performs well on various benchmarks, including MMLU, HumanEval, a Python code generation test, and MATHS, which covers middle and high school maths exams.

In a hand-graded test, Grok-1 scored a 59% in the 2023 Hungarian national high school finals in mathematics, surpassing Claude 2's 55% and GPT-4's 68%.

These numbers indicate that Grok-1 is more capable than OpenAI's GPT-3.5 but not as advanced as the latest GPT-4. xAI claims that Grok-1, despite using less training data, can compete with models requiring larger datasets and more computing power.

Grok's Training and Data Access

Grok-1 is trained using a custom stack based on Kubernetes, Rust, and JAX. It has real-time internet access, enabling it to provide up-to-date information. However, this also means it can generate incorrect or contradictory information. xAI is actively seeking human feedback to improve Grok's contextual understanding, multimodal capabilities, and adversarial robustness.

Availability of Grok

Currently, Grok's beta version is accessible to a limited number of users in the US. In the near future, it will become available to X Premium+ subscribers, costing $16 per month which translates to Rs. 1300 when subscribed from the web.

So, if you want to pay, you can enjoy the knowledge and humour Grok brings to your questions as it continues to evolve and learn.