As professionals whose business it is to integrate word, image, and type, we need to question our readiness for new uses of artificial intelligence on the horizon. I'm talking ChatGPT (now powered by the GPT-4 language model) and the impact of large language models (LLMs) on information design. A key feature of GPT-4 (released in March of 2023) is its ability to take in both images and text to generate answers in text. Earlier versions were already reasonably good at tasks such as translation and summarization, but the latest version can perform tasks such as writing business documents, academic articles, generating art, music, 3-D images, animations, and even games. GPT-4-trained by developers at OpenAI-is built on vast amounts of data scraped from the internet and elsewhere-ranging from news articles, images, artwork, and encyclopedias to social media posts and online forums. Although its comprehensiveness is impressive and better than GPT-3.5, part of its database is still loaded with bias and misinformation, raising serious concerns about the ethics that underlie its feedback.
展开▼