Up until recently, ChatGPT only “knew” about things that happened before 2022 based on it having been trained on a dataset that ended in 2021. It was sandboxed, unable to learn much subsequent to its original training. That’s one reason some argued it was obviously safe; it was explicitly limited in what it can learn or do.
Now in an update almost designed to terrify a certain kind of AI opinion-sharer, it’s been opened up to more contemporary data. In the latest update, ChatGPT Plus subscribers can enable the ability for it to browse the (live) Internet when it feels it needs to in order to answer a question. You can ask it questions like “What are today’s top news stories?” for instance.
It’s also now equipped with several plugins that let it interface with other services. Perhaps it’s only a matter of time before it pays a Taskrabbit to do something worse than solve a CAPTCHA. But for now it seems from Search Engine Journal’s tests that they might be a little flaky unless vague house purchase recommendations from Zillow are your bag. Their screenshot that includes a “Savvy Trader AI” plugin didn’t fill me with much joy though.
ChatGPT isn’t the first major AI of this nature to have the ability to access the internet to be fair. Google Bard had this as a feature from the time of launch.