As great as multimodal capabilities are, it fails to eradicate hallucinations, & lacks the ability to continually assess and amend its own output until it reaches the desired level of accuracy. So this advance doesn't fix the big problems.
Exciting times! A truly multimodal model would be amazing. I wonder when we will get the first model that can replace Siri/Assistant that we can just talk to directly?
I tried Bard today and got some bad code from it. Apart from the thumbs down reaction, there's nothing that can retrain the model.
I really want to be able to submit working code to keep refining Bard. Or at least correct it. It should be able to learn from its mistakes.