Will a photorealistic chameleon be enough to make you forget OpenAI’s worries? The new darling of Silicon Valley hopes so since the company is trying to escape a new controversy.
Can we trust AI and those who develop it? This is the question that arises, implicitly, following an interview with Mira Murati, technological manager of OpenAI, by the Wall Street Journal. Invited to present Sora, the new video creation AI recently announced by OpenAI, the manager admitted to not knowing exactly where the videos that were used to train the machine came from.
OpenAI absorbs everything
At a time when OpenAI (and the artificial intelligence industry in general) is caught between multiple controversies over data misuse, hearing an official publicly admit “not be sure» where the data used to train a machine comes from is not particularly reassuring. Explaining not wanting “go into details» on the subject, Mira Murati then explained that it was in any case data “public» or including OpenAI «owned the operating rights“.
The company later confirmed that some of the data came from Shutterstock, the well-known image bank, but it is very likely that this is just a drop in the ocean of data ingested by Sora. A new batch of videos generated by Sora actually suggests that OpenAI’s new machine has trained on other content than the rather banal videos from Shutterstock.
Multiple sources of inspiration
On Instagram, for example, we can see the very cute video of a kitten with a pirate headband riding a robot vacuum cleaner. If some 3D models still seem rudimentary, we can clearly see in this photorealistic video inspirations coming from YouTube, which for 10 years has cheerfully documented the relationships between our pet felines and our robot vacuum cleaners. Some videos of this type exceed 10 million views today.
Another video, this time showing a chameleon perched on a branch, actually seems straight out of an image bank with this characteristic sharpness and its extremely small depth of field.
While it is undeniable that the videos generated by Sora are of impressive quality, as evidenced by these colorful Niagara Falls or this elephant made of tree leaves, the vagueness surrounding their production method is perplexing. By plundering all the video content and 3D modeling available on the web without necessarily asking the opinion of the authors behind it, OpenAI exposes itself to a backlash similar to what is currently happening with photographers and designers who are “poisoning” their images to prevent them from being used for the education of AI.
Download
- Text-to-video
- Very realistic short videos
- Generative AI from Open AI
Sora has a simple goal: to take the principle of ChatGPT and Dall-E to extend it to videos. If the concept was until now not very effective, with unimpressive video results, the Open AI tool offers much more interesting precision in rendering, with in particular the possibility of controlling the movements of the camera, the rendering of objects present, etc.
Sora has a simple goal: to take the principle of ChatGPT and Dall-E to extend it to videos. If the concept was until now not very effective, with unimpressive video results, the Open AI tool offers much more interesting precision in rendering, with in particular the possibility of controlling the movements of the camera, the rendering of objects present, etc.
Source : Wall Street Journal, Instagram – OpenAI
0