On Thursday, ChatGPT maker OpenAI introduced a new text-to-video AI model named Sora, and while the announcement featured only a few initial video examples, Macquarie analysts say it is a “new breakthrough technology.”
The announcement shows several 60-second videos which OpenAI says were created “directly by Sora without modification.”
With Sora, the AI company “once again showed that its internal product developments are pushing the envelope and that its releases tend to set new bars for state-of-the-art,” analysts at Macquarie said in a note.
“We think both the length of coherent video generation, and the model's understanding of how entities exist and move in the physical world are both key breakthroughs with this technology release,” they added.
Analysts highlighted three key takeaways from the Sora release.
Firstly, it exemplifies the risks of building startups that are dependent on proprietary models for competitive advantage, given the emergence of well-funded ventures.
Secondly, the pace at which generative AI could revolutionize creative sectors, such as video production, might be faster than previously thought.
Lastly, Macaquire says the outcome of the NY Times' copyright lawsuit, “which we think is the lawsuit to watch to understand how courts could apply copyright law to generative AI companies,” will likely influence the speed at which AI models like Sora, which presumably require substantial media data for training, are developed.