Technology Vision 2023 | When Atoms meet Bits #TechVision When OpenAI revealed ChatGPT—a powerful new driving controversy. Arguments over fairness in art 169 chatbot—in late 2022, people clambered to test it. competitions, the ethics of mimicking artists’ styles, They asked complicated, open-ended questions, legal risks, and the impact on people’s livelihoods requested poems and essays with obscure topics have 昀氀ared. And yet, while relevance to the art and styles, and got impressive and sometimes industry is clear, businesses in other 昀椀elds may still humorous results: in one case, instructions for see it as mere novelty—and they’re making a mistake. removing a peanut butter sandwich from a VCR 170 written in the style of the King James Bible. This generated content is part of one of the biggest step changes in the history of AI: the introduction of Then, not even four months later, OpenAI released pretrained models with remarkable task adaptability, GPT-4, the greatly anticipated next generation of its which will revolutionize how and where enterprises AI software, which is even more creative, and can across industries use AI. respond to both image and text inputs and even 171,172 more nuanced instructions. It began with a landmark innovation in AI model 173 architecture by Google researchers in 2017. Since And before all this, the internet was 昀氀ooded with then, tech companies and researchers have been AI-generated art. Text-to-image generators like supersizing AI, increasing the size of models by Stability AI’s Stable Diffusion, Midjourney, and 10,000 times and the size of training sets too. The OpenAI’s DALL-E 2, stunned people by responding result: powerful, pretrained models, often called to written prompts with photorealistic images. “foundation models,” that offer unprecedented “Two capybaras waltzing in the style of Monet” adaptability within the domains they’re trained on, would get you pretty much that. be it language, images, or the structure of proteins. With this adaptability, foundation models can In fact, this wave of AI-generated content is so complete a wide variety of tasks without needing advanced—and moving so quickly—that it’s task-speci昀椀c training. What’s more, companies