In a March 29 blog post, Elon Musk announced the release of an upgraded version of the Grok chatbot next week. This version will be available in advance to early testers and existing Grok users on the social media platform.
Grok-1.5 is an enhanced version of Grok-1, improving inference and can handle contexts up to 128,000 words long. xAI analysis also shows that chatbots also demonstrate superior performance in tasks related to coding and mathematics.
However, Grok-1.5 is still inferior to Google's Gemini Pro 1.5 and OpenAI's GPT-4. According to research, Grok-1.5 only scored 50.6% on the MATH benchmark and 90% on the GSM8K benchmark, these are two benchmarks that evaluate math solving ability at levels from elementary to high school.
Additionally, this virtual assistant only scored 74.1% on the HumanEval benchmark, used to evaluate code creation and problem-solving abilities.
Analysis of AI Chatbot. (Source: xAI)
xAI added that Grok-1.5 is built on a custom distributed training platform based on JAX, Rust and Kubernetes, making it easy for the xAI team to generate ideas and train new architectures at scale. .
Earlier this month, xAI took a significant step forward by open sourcing the Grok-1 code base. The decision stems from a legal case initiated by Musk against OpenAI, the organization he once co-founded. Musk accused OpenAI of deviating from its initial commitment to prioritize open source development over shareholder interests.
The released source code is a rough base model test from Grok-1's pre-training phase, completed in October 2023. This means that the model is not fine-tuned for any specific application, such as conversation.