The successors of GPT-3 might achieve that. They should be able to operate in both text and image/video domain, having all these modalities as both input and output, have a longer context window and be up to date with the recent news/publications in the domains of interest of the user.

But fundamentally it shows it can be done. It's a general language interface to a massive trove of knowledge and can perform a yet undetermined number of symbolic manipulation and reasoning tasks. A future version could be VR enabled, creating a holodeck like experience - something that could prove to be very addictive.