How to See the Future with DALL-E 3? Earlier this month, I discussed how you may use commercial versions of frontier AI products to travel back in time. Let’s take a look at another of those items today: the text-to-image generator DALL-E. The most recent version reveals a lot about how OpenAI is thinking about products, regulations, and safety as its core technology evolves.
How to See the Future with DALL-E 3?
ChatGPT is getting all of the buzz these days. However, six months before OpenAI’s chatbot came and moved the tech world’s collective emphasis to generative artificial intelligence, the company published another tool that alluded to comparable possibilities. DALL-E 2, which I first saw in June of last year, fascinated me from the start.
However I rapidly discovered the limits of my creativity, and despite my best efforts to follow the instructions in books aimed to help me become a better illustrator, I never got very far. After a few decades, I found myself able to construct entire worlds just by typing words into a box. It seemed magical in a way that the tech sector frequently promises but rarely delivers.
In the 18 months after DALL-E 2’s release, problems concerning copyright, licences, and what, if anything, the creators of text-to-image generators like DALL-E owe the artists whose work their models are trained on have muddied the picture. Getty Images sued Stable Diffusion earlier this year, alleging that the company’s algorithm was unlawfully trained on its photographs. Similar cases appear to be on the way. Meanwhile, Adobe exhibited a different road ahead by developing its own Firefly picture generator using only licenced images, and it has stated that it will compensate creators whose work was utilised in the training process.
Despite the legal and ethical uncertainties surrounding text-to-image generators, the field has advanced swiftly. Despite being available only on Discord, Midjourney, which began shortly after DALL-E 2, has 15 million users and generates hundreds of millions of dollars in annual income with its own picture generator. Last October, Stable Diffusion has 10 million users. As they grew, the quality of their photographs improved rapidly, but DALL-E’s stayed stationary, although being amazing by pre-2022 standards.
DALL-E 3 landed on Thursday. After a brief period in public testing, the next generation of OpenAI’s picture generator is now available to commercial customers and ChatGPT Plus members. (A free version is also available through the Bing Image Creator.)
For the past few weeks, I’ve been utilising DALL-E 3 via Bing and ChatGPT. On Wednesday, I was also briefed on the updated version by Gabriel Goh, an OpenAI research scientist who helped design the new model and Sandhini Agarwal, the company’s AI policy manager.
Also Read: How to Use ChatGPT Voice to Learn a Language?