The xAI startup has introduced its artificial intelligence known as Grok. This product not only answers questions in any field but also suggests what questions users should ask. The AI mentioned for responding to queries has a bit of a playful and rebellious nature, so if you’re not in the mood for humor, it’s recommended not to use it.
According to xAI’s official website, Grok’s unique advantage is its access to real-time data from the X platform (formerly Twitter). This tool also provides answers to specific questions that most AI chatbots dismiss.
Grok is still in the experimental phase, and the best thing to do is to practice with this chatbot. In fact, Grok’s performance is expected to improve with more experience over time.
xAI states that it aims to create new AI tools that assist humanity in understanding and knowledge. They outline their goal for creating Grok as follows:
- Improving research and innovation: They want Grok to act as a powerful research assistant for everyone. Their AI helps individuals quickly access the information they need, process data, or receive new ideas.
- Gathering feedback and ensuring that AI tools maximize benefits for all of humanity: They believe that designing AI tools for people with any background and political perspective is vital. They also intend to enhance users’ capabilities with such tools while complying with the law.
xAI’s ultimate goal in creating Grok is to help humanity gain a better understanding of the world.
Birth of Grok-1
The main engine of Grok, the large language model Grok-1, was developed over four months and several versions were produced during this time.
xAI initially trained its first large language model (Grok-0) with 33 billion parameters. This model approaches capabilities similar to LLaMA 2, but its training resources are nearly half of the competition. The progress made on the initial Grok model over the past months culminated in the development of Grok-1, which is significantly more powerful and scored 63.2% and 73% in HumanEval and MMLU metrics for reasoning and coding, respectively.
On average, Grok-1 outperforms most current models in its computational class, including GPT 3.5 and Inflection-1, in benchmarks like GSM8k, MMLU, HumanEval, and MATH. Only models like GPT-4, trained with significant amounts of training data and computational resources, continue to outperform Grok, highlighting xAI’s rapid progress in training large language models with exceptional performance.
Early Access to Grok
xAI currently offers initial access to Grok for a limited number of users in the United States to gather initial feedback on its AI’s performance. This step helps the company enhance various features and address potential issues before widespread release. You can also join the Grok waitlist through this link.
Now, it remains to be seen how Grok will perform in competition with rival chatbots like ChatGPT, Google BERT, and Microsoft’s Bing Chatbot.
