0
0
Elon Musk Announces Upgraded Version of xAI Chatbot Grok
Elon Musk recently disclosed that an enhanced iteration of his artificial intelligence firm xAI’s chatbot Grok is set for release in the upcoming week. This announcement was made through Musk’s social media post on March 29, which followed xAI’s revelation of Grok-1.5 in a blog post. The upgraded AI chatbot will initially be available to early testers and existing Grok users on the social media platform.
In addition to this development, Musk also hinted at the continuous advancement of Grok 2, which he expects will exceed current AI standards across various domains.
Grok-1.5
The latest version, Grok-1.5, represents a significant advancement over the Grok-1 AI model and boasts improved reasoning capabilities along with a context length of 128,000 tokens. Analysis conducted by xAI indicates notable enhancements in the performance of this advanced chatbot, particularly in tasks related to coding and mathematics. However, it falls short in comparison to Google’s Gemini Pro 1.5 and OpenAI’s GPT-4.
“Grok-1.5 achieved a 50.6% score on the MATH benchmark and a 90% score on the GSM8K benchmark, both of which cover a wide range of math problems from grade school to high school competitions. Furthermore, it attained a 74.1% score on the HumanEval benchmark, evaluating code generation and problem-solving capabilities.”
Moreover, Grok-1.5 has the ability to extract information from significantly longer documents, enabling it to handle extended and more complex prompts while maintaining its proficiency in following instructions as the context window expands. The development team at xAI utilized a custom distributed training framework based on JAX, Rust, and Kubernetes to create Grok-1.5, facilitating the prototyping of ideas and training of new architectures at scale with minimal effort.
Grok Goes Open-source
Earlier this month, xAI took a pivotal step by open-sourcing the base code of Grok-1 in response to legal action initiated by Musk against OpenAI, an organization he co-founded. Musk alleged that OpenAI had shifted its focus away from prioritizing open-source model development in favor of shareholder interests.
Subsequently, xAI released the raw base model checkpoint from the Grok-1 pre-training phase, concluding in October 2023. This version of the model is not fine-tuned for any specific application, such as dialogue, and is licensed under the Apache License 2.0.
Image/Photo credit: source url
