When artificial intelligence creates masterpieces to order

A new artificial intelligence program created by OpenAI (often insane) creates amazing images based on descriptions.

Photo by DALL · E 2 Mona Lisa is seen eating cake at the end of the world @Twitter screenshot

Do you know DALL · E 2? Otherwise it is normal. This is a brand new technology created this year. But in recent weeks, he has started buzzing on social networks. The key to its success: this artificial intelligence (AI) creates realistic images from arbitrary sentences without human intervention. If we want to summarize, we can say that it is a question of depicting words. A very common idea but the quality is so high that we are amazed to see these creations which are all masterpieces.

Babylon version Basketball astronaut with Dali and cat

To give you an example, let’s take the following sentence: “The hanging garden of Babylon in the middle of a city, in the style of DaliIt makes no sense but based on the various elements that compose it, AI will create a whole new image. It will look for photos of the Garden of Babylon, representing a city, and it is inspired by Dali’s creation. Mixing everything together gives a painting worthy of a Catalan painter, although he apparently never made it. If you want a canvas that represents a fox in the style of Van Gogh’s The Starry Night or Claude Monet.

The idea is infinitely undeniable. You can ask DALL · E 2 to create a presentation for astronauts playing basketball with a cat in the style of a children’s book or a minimalist or even a watercolor. AI can also capture an existing image, such as Vermeer’s famous Girl with a Pearl Earring, and create multiple variations of it. Each time, the results are more than credible. No worries about making beautiful book covers or flashy posters. The machine does this for you (regardless of the artist’s job risk, but this is another discussion).

Avoid drift to increase interest in the program

To achieve this result, the OpenAI company created the first DALL · E in January 2021. Already, AI has been able to generate such results from sound. A strategy whose nickname is “expansion”. One year later, DALL · E 2 multiplied a resolution by four and pushed the bar a little further, resulting in a much more reliable image. The reality definitely goes to a higher level and the correspondence with the legends is much sharper.

If you’re worried about AI creating violent images by people with bad intentions, OpenAI says it can trust you. The company has actually created an algorithm to remove such content, at least explicitly. The image that would emerge from such a sentence would be very metaphorical. DALL · E 2 avoids using the real faces of individuals, especially celebrities, to avoid the same kind of abuse as Dipfeck.

Our hope is that DALL · E 2 will allow people to express themselves creatively. DALL · E 2 helps us understand how advanced AI systems can see and understand our world, which is important for our goal to create AI that benefits humanity.“, Explains OpenAI.

DALL · E 2’s “secret language”

Obviously, DALL · E 2 is not perfect. He sometimes misrepresents the ideas present in a sentence, especially when it comes to neologism. This was reported by The Independent, which tells the story of Giannis Daras, a doctoral student in computer science in Austin. The latter asked the farmers to make a picture while talking to the vegetables when the word “Vikuts” was introduced (which does not exist). Result: The figure shows the dishes with vegetables. Upon closer inspection, the University of Texas discovered that DALL · E 2 was creating its own vocabulary (which is as unreasonable as the specific sentences submitted to it). In its language, “Apoploe vesrreaitais” means “bird” for example. “The discovery of Dale-2 language poses many interesting security and interpretive challenges“Texan computer scientists are surprised to follow the study published. Their conclusion still needs to be the subject of additional research to confirm that AI has created a great language on its own!

If you want to test yourself, OpenIA allows you to join a waiting list for submitting a sentence. The results can then be posted on social media. Many internet users have tried this with more bizarre phrases than each other (“Steve Jobs as a happy meal toy”, “Darth Vader’s eating fry” etc.) before sharing pictures on their Twitter or Instagram accounts.

Leave a Comment