In one of GPT4's release demos, it is shown being fed a photograph of a hand-drawn web page. It then image analyses that, works out that it's a picture of a web

In one of GPT4's release demos, it is shown being fed a photograph of a hand-drawn web page. It then image analyses that, works out that it's a picture of a web page and turns it into a working web page (https://youtube.com/live/outcGtbnMuQ?feature=shares&t=972 https://youtube.com/live/outcGtbnMuQ?feature=shares&t=972). So here's the thing; what in that list of non-codegen tasks is more complicated than that which GPT can already automate?