OpenAI has announced a major upgrade to the technology that underpins ChatGPT, the seemingly magical online tool that professionals have been using to draft e-mails, write blog posts and more. If you think of ChatGPT as a car, the new language model known as GPT-4 adds a more powerful engine. The old ChatGPT could only read text. The new ChatGPT can look at a photo of the contents of your fridge and suggest a dinner recipe. The old ChatGPT scored in the 10th percentile on the bar exam. The new one was in the 90th. In the hours since its release, people have used it to create a website from a hand-drawn sketch or look through a dating website for an ideal partner.
This is the fun part of unleashing a powerful language model to the public – the honeymoon period. But what are the long-term consequences? OpenAI (once again) has not disclosed the datasets it used to train GPT-4, so that means researchers cannot scrutinise the model to determine how it might inadvertently manipulate or misinform people. More broadly though, it ushers in a new era of hyper-efficiency, where professionals will have to work smarter and faster – or perish.