This paper explores the capabilities of an early version of OpenAI's GPT-4, highlighting its potential as a step towards artificial general intelligence (AGI). The authors investigate GPT-4's performance on a wide range of tasks, including language, vision, coding, mathematics, and interdisciplinary composition. They demonstrate that GPT-4 can solve novel and complex problems across various domains without special prompting, often achieving human-level performance or surpassing previous models like ChatGPT. The paper discusses GPT-4's strengths and limitations, emphasizing its generative and interpretative abilities. It also addresses the challenges and societal implications of advancing AI systems with broad capabilities. The authors conclude by reflecting on the future directions for AGI research, including the need for new paradigms beyond next-word prediction.This paper explores the capabilities of an early version of OpenAI's GPT-4, highlighting its potential as a step towards artificial general intelligence (AGI). The authors investigate GPT-4's performance on a wide range of tasks, including language, vision, coding, mathematics, and interdisciplinary composition. They demonstrate that GPT-4 can solve novel and complex problems across various domains without special prompting, often achieving human-level performance or surpassing previous models like ChatGPT. The paper discusses GPT-4's strengths and limitations, emphasizing its generative and interpretative abilities. It also addresses the challenges and societal implications of advancing AI systems with broad capabilities. The authors conclude by reflecting on the future directions for AGI research, including the need for new paradigms beyond next-word prediction.