Tech

Innovative AI is changing everything. But what remains when the hype is gone?

The big breakthrough behind the new models is how the images are created. The first version of DALL-E used an extension of the technology behind Language model of OpenAI GPT-3, which creates images by predicting the next pixel in the image as if they were words in a sentence. This works, but not well. “It was not a magical experience,” says Altman. “It’s amazing how effective it is.”

Instead, DALL-E 2 uses something called a diffusion pattern. Diffusion models are neural networks trained to clean images by removing the pixel noise that the training adds. This process involves taking an image and changing a few pixels in it at once, over several steps, until the original image is erased and you are left with nothing but random pixels. “If you do this thousands of times, it will end up looking like you just pulled an antenna,” says Björn Ommer, who works on artificial intelligence at the University of Munich in Germany and helped build it. -ten from my TV—just snow.” Diffusion pattern now powers the Stable Diffusion.

The neural network is then trained to reverse that process and predict what a less pixelated version of a given image will look like. The end result is that if you feed a diffuse model a bunch of pixels, it will try to produce something a bit cleaner. Plug in the cleaned image again and the model will generate something cleaner. Do this enough times and the model can take you from a snowy TV to a high-resolution picture.

The AI ​​art generator never works exactly the way you want it to. They often produce hideous results that can resemble distorted pre-existing artwork. In my experience, the only way to really make work look good is to add a description at the end with a style that looks aesthetically pleasing.

~Erik Carter

The trick with text-to-image models is that this process is guided by the language model trying to match the prompt to the image the diffusion model is generating. This pushes the diffusion model to images that the language model considers appropriate.

But the models don’t pull the links between text and images out of the air. Most of today’s text-to-image models are trained on a large dataset called LAION, which contains billions of pairs of text and images pulled from the internet. This means that the image you get from the text-to-image model is a distillation of the world as it is presented online, distorted by prejudice (and pornography).

One last thing: there is a small but very important difference between the two most popular models, the DALL-E 2 and the Diffuse Stable. DALL-E 2’s diffusion model works on full-sized images. Stable diffusion, on the other hand, uses a technique called latent diffusion, invented by Ommer and his colleagues. It works on compressed versions of images encoded in a neural network in what is known as latent space, where only the essential features of the image are retained.

This means that Stable Diffusion requires less computational mechanisms to operate. Unlike DALL-E 2 which runs on OpenAI’s powerful servers, Stable Diffusion can run on personal computers (well). Much of the explosion of creativity and rapid growth of new applications is due to Stable Diffusion being both open source—developers are free to change it, build on it, and monetize it. —light enough for anyone to run at home.

Redefining creativity

For some, these models are a step forward artificial intelligence, or AGI—an overblown buzzword referring to a future AI with general-purpose or even human-like capabilities. OpenAI has been clear about its goal of achieving AGI. For that reason, Altman doesn’t care that DALL-E 2 is currently competing with a range of similar tools, some of them free. “We are here to create AGI, not an image generator,” he said. “It will align with the broader product roadmap. It’s a small element of what AGI will do.”



Source by [author_name]

news7h

News7h: Update the world's latest breaking news online of the day, breaking news, politics, society today, international mainstream news .Updated news 24/7: Entertainment, Sports...at the World everyday world. Hot news, images, video clips that are updated quickly and reliably

Related Articles

Back to top button