In short Everyone agrees that text-to-image models are here to stay, although opinions are divided on AI-generated art.
Some artists are captivated by the ability to create entirely new digital images using text prompts and see it as a new tool for being creative. Other artists, however, hate technology and believe it will take away their jobs and devalue their work.
A machine can be trained to recreate the style of a particular artist and surpass human artists, as concept artist RJ Palmer told the BBC. “Right now, if an artist wants to copy my style, they can spend a week trying to reproduce it. It’s a person who spends a week to create one thing. With this machine, you can produce hundreds of them a week /”
The AI ”directly steals their essence in a way”, and artists are currently powerless to stop it from happening. Developers train these models by feeding them with a large database of images pulled from the Internet. It’s not too surprising that some of their work ends up in a model’s training dataset.
The creator of Stable Diffusion, a popular open model taking the internet by storm, however, said he doesn’t believe AI will take artists’ jobs. Excel “didn’t put accountants out of work, I always pay my accountants,” said Emad Mostaque.
He said the tool will provide new jobs for artists: “It’s a sector that’s going to grow massively. Earn money with this sector if you want to earn money, it will be a lot more fun.”
Jason Allen, who controversially won a state art fair with a digital image, once said, “The art is dead, man. It’s over. The AI won. The humans lost. .”
Cruise is expanding its AI robotaxi service
Self-driving car company Cruise will launch its self-driving taxi service to cities in Texas and Arizona by the end of this year.
Co-founder and CEO Kyle Vogt told TechCrunch that the company plans to operate a small fleet of self-driving vehicles on the roads of Austin, Texas, and Phoenix, Arizona “in the next 90 days and before the end of 2022”. Cruise has launched its first human-driverless robotaxi service in San Francisco, California.
The service only operates in a few selected areas late at night from 10:00 p.m. until 5:30 a.m. to avoid rush hour traffic. Not everyone can call a car, however, only a small group of pre-selected passengers can. The waiting list for members of the public to consider and join is open.
Vogt said Cruise also hopes to start driving newly designed Origin vehicles next year. These square cars will have no steering wheel or pedals and will be fully automated. “Looking at 2023, next year things get really interesting on the growth side,” he said.
“Thousands of AV vehicles will roll out of the General Motors factory, including the first Origins. We will use them to light up many other markets and begin to generate meaningful revenue in those markets.”
Will the AI community be stuck with transformers?
A creator of the popular AI library, PyTorch, has warned that the current trend of optimizing hardware for transformer models will make it harder for new architectures to succeed.
Transformers were first used in natural language processing and are the source of the most powerful generative models yet capable of creating text and images. They’ve been used in all sorts of applications, from gaming to drug design. Hardware companies like Nvidia are optimizing their chips to speed up transformer-based designs, which could hamper innovation in the future.
Soumith Chintala, who helped build PyTorch, told Business Insider that he hopes a different kind of pattern will emerge.
“We’re in this weird hardware lottery. Transformers came about five years ago, and another big thing has yet to see the light of day. So companies may be thinking ‘we should just optimize hardware for transformers”. It follows then that going in any other direction is much more difficult.
“Different processor architectures will not work as efficiently on current and future chips, and may deter developers from offering other types of designs.
“It will be much more difficult for us to try other ideas if hardware vendors end up making accelerators more specialized compared to the current paradigm,” Chintala warned. ®