Read the original document by opening this link in a new tab.
Table of Contents
Abstract
1 Introduction
2 Scope and Limitations of this Technical Report
3 Predictable Scaling
3.1 Loss Prediction
3.2 Scaling of Capabilities on HumanEval
4 Capabilities
Exam GPT-4
Table 1. GPT performance on academic and professional exams
Table 2. Performance of GPT-4 on academic benchmarks
Summary
This technical report presents GPT-4, a large multimodal model capable of processing image and text inputs and producing text outputs. GPT-4 is a Transformer- based model pre-trained to predict the next token in a document. It exhibits human-level performance on various professional and academic benchmarks, including passing a simulated bar exam with a score around the top 10% of test takers. The report discusses the development of GPT-4, its capabilities, limitations, and safety properties. The project focused on building a deep learning stack that scales predictably and improving the understanding of natural language text. GPT-4 outperforms existing language models and state-of-the-art systems on various benchmarks. The report also highlights the potential societal impact and safety challenges posed by GPT-4.