• by Laurie Sullivan  @lauriesullivan, November 6, 2023

Elon Musk, one of the early investors in OpenAI's technology, released his own version of ChatGPT during the weekend. He claims the prototype is already superior to ChatGPT 3.5 in several benchmarks. 

Grok -- the first product of Musk's xAI company -- is being tested with a limited group in the United States.

The chatbot is supported with data from Musk's X social network, giving it an advantage of real-time knowledge of the world. It is designed to answer questions with a "bit of wit" and has a "rebellious streak," the xAI team said.

The technology was announced in July, supported by a group of executives at the forefront of AI from Google's DeepMind, Microsoft and Tesla, among others. 

"Grok is an AI modeled after the Hitchhiker’s Guide to the Galaxy, so intended to answer almost anything and, far harder, even suggest what questions to ask!" the AI team said. The post explains how the technology aims to assist in the “pursuit of understanding.”

The engine powering Grok is Grok-1, xAI’s frontier large language model (LLM), which the team developed during the last four months. The prototype LLM, Grok-0, was trained with 33 billion parameters.

In the past two months, the team made significant improvements in reasoning and coding capabilities. Grok-1 now achieves a 63.2% score on the HumanEval coding task and 73% on MMLU.

To get a better understanding of the achievements, the team conducted a series of evaluations using standard machine-learning (ML) benchmarks designed to measure math and reasoning abilities.

GSM8k: Middle-school math word problems

MMLU: Multidisciplinary multiple-choice questions

HumanEval: Python code completion task

MATH: Middle-school and high-school mathematics problems written in LaTeX

The group gave Grok access to search tools and real-time information, but as with all the LLMs trained on next-token prediction, the model can still generate false or contradictory information. The goal is for the technology to achieve reliable reasoning.

Musk has been vocal about his opposition to the way OpenAI trained its ChatGPT models. He was not the only contributor to OpenAI’s funding in 2015.

Global companies such as Amazon Web Services, Infosys and Tesla Motors, along with Peter Thiel and others, also contributed to the $1 billion investment.

At the time, the nonprofit OpenAI focused on research aimed at advancing digital intelligence to benefit humanity. 

The group intended to become the governing body of AI, overseen by research scientists and developers who have the expertise to determine the possibilities and potential dangers of initiatives in artificial or machine intelligence.

Other investors included Reid Hoffman and Jessica Livingston. The group's other founding members are leading research engineers and scientists, including Trevor Blackwell, Vicki Cheung, Andrej Karpathy, Durk Kingma, John Schulman, Pamela Vagata and Wojciech Zaremba.

Pieter Abbeel, Yoshua Bengio, Alan Kay, Sergey Levine and Vishal Sikka are advisors to the group. OpenAI's co-chairs are Y Combinator's Sam Altman and Musk, the company said Friday.

Microsoft's contribution, which is worth billions, did not come until later -- building the AI technology into much of its services and software, including Microsoft Bing, the company’s search engine, and Microsoft Advertising for advertisers.   

In late October, U.S. President Joe Biden signed an executive order related to AI with the goal of finding a balance in the industry by creating an early set of guardrails supported by legislation and global agreements.

The order created an initial step to ensure that AI would become trustworthy and helpful rather than deceptive and destructive. It aims to guide how AI is developed so companies can profit without putting public safety in jeopardy.