Open AI shared the results of their research, which I can not call anything other than magic – DALL · E, a new neuron, a continuation of the GPT-3 idea on transformers, but this time for generating images from text.
I often write here about fantasy, they say, I fed Harry Potter to the neuron and received illustrations of all the scenes of the book – it seems this is no longer a fantasy, but they still don’t give me anything to dig deeper.
DALL · E neuron with 12 billion parameters, trained in picture-text pairs, its tasks:
- Synthesize pictures by text description
- Draw pictures with a part at the input, taking into account the text description
Open AI has already teased some things in this area before, and finally it has reached such a level that the jaw is falling off, look at the examples that I have attached, at the top of the text that was given at the input.
I’m sure she won’t be allowed to play for everyone yet.
I foresee this research will greatly affect many areas and industries, as the applications are endless.
More details here: https://openai.com/blog/dall-e/
Own AI Designer
A fiction that has already become a reality. You tell the computer (in words!) what you want to see in a picture, and it draws it.
One more small step for GPT-3 and a huge leap for the design industry. You describe a photo/illustration in words and voila (so far 256×256). In short, there will be more good design around in the next 10-15 years. Designer’s work will become much more interesting (a lot of routine will go away).
This is the client’s paradise: “play with fonts”, “enlarge the logo”, “highlight in red” – all this is readily done by a mentally balanced machine.
On the other hand unusual illustrations for children’s books, for example, are also welcome.
What profession to master in order not to be left behind in 10 years?