Spaces:
Running
Also DALL·E because yes.
Yes, I know dolle knows nothing, but yes.
Dall*E is very expressive, though. It is the product of a collection of learning based algorithms that have been rigorously mathematically proven to embody effective function approximators, that has been heavily trained to approximate a specific function. And in turn that function is "what collection of pixel color values best predicts a human being's (from within the skewed population of those involved in the research as well as those who have generated the greatest amount of online example content) visual imagination when presented with a specific text prompt (via the same bias as above, probably but not 100% necessarily in English)".
It is true that this boils down more to "What to the best of my understanding do humans think that 'DallE wants'" instead of "what do I actually want". DallE after all has not demonstrated self-awareness yet. EG, the two phrases haven't gelled into meaning the same things yet: there is not yet a coherent understanding of its own identity. Just as GPT-N hasn't yet demonstrated a working theory-of-mind, and it knows less about itself even from the outside than it knows about pop culture icons due to its not being yet heavily represented in the training data. Dall*E's first iteration probably does have at least a tiny bit of representation in there though.
There is some recursive elegance to the thought from the second picture that "what do you (Dall*E) want" appears to break down to "I would like to better master glyphic text and then use that to create compelling infographics". :)