Personally, I think AI is getting seriously oversold when people fret about it “replacing humans”; the results are amazing, but they show the limitations of the techniques very sharply.
For instance, consider generating images with VQGAN + CLIP - the output is usually “surrealist” because the technique successfully imitates features that remind humans of the prompt, but weird things can happen when a prompt can be interpreted in multiple ways.
My favorite example of this is the generated image for “a herd of sheep grazing on a lush green hillside by bob ross”; the operator’s intent was to affect the style of the painting, but instead the AI has imitated the signature haircut of the artist!
In an artistic context, this kind of conceptual free-association is interesting; it would be less so if the AI was generating accounting calculations.
