OpenAI’s new GPT-4 AI model is a remarkable machine learning system that has been powering everything from virtual volunteers for the visually impaired to an improved language learning bot in Duolingo. GPT-4 is quite different from previous versions like ChatGPT and GPT-3.5 in five main ways.
Firstly, GPT-4 is “multimodal,” which means that it can understand more than one “modality” of information. Unlike ChatGPT and GPT-3, which were limited to text, GPT-4 can process images to find relevant information. This means that GPT-4 can describe what is in a picture, but its understanding goes beyond that. GPT-4 describes the pattern on a dress, identifies a plant, explains how to get to a machine at the gym, translates a label (and offers a recipe), reads a map, and performs a number of other tasks. However, one may not always know whether a dress is the right outfit for an interview.
Secondly, GPT-4 is much harder to trick. For instance, it has been trained on lots of malicious prompts, which makes it much better than its predecessors on “factuality, steerability, and refusing to go outside of guardrails.” The new model was “unprecedentedly stable,” and OpenAI applied the lessons learned from GPT-3.5 to GPT-4, resulting in fewer surprises.
Thirdly, GPT-4 has a longer memory. While the old version of ChatGPT and GPT-3 had a limit of 4,096 “tokens” that they could keep “in mind,” GPT-4 has a maximum token count of 32,768. This means that in conversation or in generating text, it can keep up to 50 pages or so in mind, which is enough for an entire play or short story.
Fourthly, GPT-4 is more multilingual. It has demonstrated that it can answer thousands of multiple-choice questions with high accuracy across 26 languages, from Italian to Ukrainian to Korean. It’s best at the Romance and Germanic languages but generalizes well to others. This is a promising testing of language capabilities, which speaks to the possibility of GPT-4 being much more friendly to non-English speakers.
Finally, GPT-4 is much faster than its predecessors. It can generate text at lightning speed, making it highly efficient for businesses that require large volumes of content. GPT-4’s amazing capabilities, coupled with its ability to generate text at such a fast pace, meaning that the technology can be used for a broad range of applications, from customer service chatbots to virtual assistants to content creation tools for businesses.