GPT-4: Definition and Examples

GPT-4 (Generative Pre-trained Transformer 4) is the fourth generation of OpenAI's language model, launched in March 2023. It represents a major qualitative leap over GPT-3.5, with significantly improved reasoning abilities, better understanding of nuances, and a marked reduction in hallucinations. It is the first model in the GPT series to be officially multimodal, accepting both text and images as input.

Unlike its predecessors, GPT-4 excels at complex tasks requiring multi-step reasoning, fine-grained document analysis, or deep contextual understanding. It has achieved scores comparable to top human candidates on standardized exams such as the US Bar Exam (90th percentile) and AP exams. Its extended context window (up to 128,000 tokens in the Turbo version) allows it to process long documents in a single request.

In prompt engineering, GPT-4 stands out for its ability to follow complex, structured instructions with high fidelity. It responds better to detailed prompts with precise constraints, specific output formats, and examples. Techniques such as chain-of-thought, few-shot learning, and system prompting achieve their full potential with this model.

GPT-4 is accessible via OpenAI's API and ChatGPT Plus. It has since been supplemented by successive versions (GPT-4 Turbo, GPT-4o) that improve speed, cost, and multimodal capabilities while retaining the reasoning foundation that made GPT-4's reputation.