GPT-4 is a large multimodal model developed by OpenAI, designed to accept both text and image inputs and generate text outputs. Released on March 14, 2023, it serves as a successor to GPT-3.5, offering enhanced reasoning capabilities and improved factual accuracy. The model was trained on a combination of publicly available data and licensed third-party information, followed by refinement using Reinforcement Learning from Human Feedback (RLHF).
The model demonstrates proficiency across a variety of professional and academic benchmarks, including scoring in the 90th percentile on a simulated Uniform Bar Exam. Its multimodal functionality enables users to submit visual data—such as charts, diagrams, and photographs—for the model to analyze, describe, or use in complex problem-solving. This makes it applicable for tasks ranging from technical documentation analysis to creative collaboration.
OpenAI implemented a safety-focused development cycle for GPT-4, incorporating six months of iterative alignment and adversarial testing involving over 50 experts in fields such as cybersecurity and international security. While these measures significantly reduced the likelihood of harmful or incorrect responses compared to previous models, the model still retains typical limitations of large language models, including hallucinations and a knowledge cutoff.