Sparks of Artificial General Intelligence: Early experiments with GPT-4

• DOI: 10.48550/arxiv.2303.12712
• Publication Date: 2023-01-01
• Authors: Sébastien Bubeck; Varun Chandrasekaran; Ronen Eldan; Johannes Gehrke; Eric Horvitz; Ece Kamar; Peter Lee; Yin Tat Lee; Yuanzhi Li; Scott Lundberg; Harsha Nori; Hamid Palangi; Marco Túlio Ribeiro; Yi Zhang
• Keywords: Predictive coding; Artificial General Intelligence

Artificial intelligence (AI) researchers have been developing and refining large language models (LLMs) that exhibit remarkable capabilities across a variety of domains tasks, challenging our understanding learning cognition. The latest model developed by OpenAI, GPT-4, was trained using an unprecedented scale compute data. In this paper, we report on investigation early version when it still in active development OpenAI. We contend (this of) GPT-4 is part new cohort LLMs (along with ChatGPT Google's PaLM for example) more general than previous AI models. discuss the rising implications these demonstrate that, beyond its mastery language, can solve novel difficult tasks span mathematics, coding, vision, medicine, law, psychology more, without needing any special prompting. Moreover, all GPT-4's performance strikingly close to human-level performance, often vastly surpasses prior such as ChatGPT. Given breadth depth capabilities, believe could reasonably be viewed (yet incomplete) artificial (AGI) system. exploration put emphasis discovering limitations, challenges ahead advancing towards deeper comprehensive versions AGI, including possible need pursuing paradigm moves next-word prediction. conclude reflections societal influences recent technological leap future research directions.