OpenAI CEO Sam Altman speaking in February Stephen Shankland/CNET
The most popular artificial intelligence technology, OpenAI GPT, received a major update on Tuesday and is now available in the premium ChatGPT chatbot.
According to OpenAI, the new GPT-4 can generate much longer text lines, react when people give it images, and is designed to better avoid the AI pitfalls seen in the old GPT-3.5. For example, in the bar exams that lawyers must pass to practice law, GPT-4 is in the top 10% of scores compared to the bottom 10% for GPT-3.5, the company said.
GPT stands for Generative Pretrained Transformer, referring to the fact that it can generate text on its own and uses an artificial intelligence technology called a transformer, which was pioneered by Google. He is trained to work with vast amounts of data collected from the Internet, learning mathematical methods to find patterns and reproduce styles.
GPT-4 – Multimodal AI
OpenAI has been making its GPT technology available to developers for years, but ChatGPT offers a simple interface that has sparked an explosion of interest, experimentation, and concerns about the technology’s shortcomings.
Last week, information was already leaked, according to which GPT-4 will allow you to generate text from video. Andreas Braun, CTO of Microsoft Germany, mentioned “multi-modal models that will offer completely different experiences, such as video.”
Because GPT-4 is a multi-modal AI, use cases can go beyond text, unlike previous models. Holger Kenn, director of business strategy for Microsoft Germany, said that AI can transform text into videos, music and images.
The text-to-video generator is not a completely new concept, as tech giants like Meta and Google already have templates like this. Meta has Make-A-Video and Google Imagen Video. Both of them use AI to create videos from user data. However, Meta and Google technologies remain in the research phase and not in the public domain, which puts OpenAI in an ideal position to stay ahead of the generative AI niche.