Last month, OpenAI unveiled Sora, an AI model that creates high-quality short Text-to-Video videos, with clips up to 1 minute long. This innovation has gained widespread attention and raised concerns about its potential impact. However, OpenAI has emphasized that Sora is still in its early stages and requires thorough risk assessment before being widely released to the public.
Recently, The Wall Street Journal conducted a special interview with Mira Murati, the COO of OpenAI, regarding the plans for Sora. Mira revealed several interesting points from the plan. According to Mira, Sora is expected to be publicly available later this year, as it is currently under the supervision of the Red Team, responsible for evaluating all risks associated with the AI’s public release.
Furthermore, the discussion shifted towards the energy and processing costs involved in creating video clips with Sora, which are likely to be significant. While Mira did not provide direct figures, she mentioned that the current costs are higher than producing images with Dall-E or engaging in chat conversations with ChatGPT. The team’s goal is to optimize cost and energy usage to be in line with similar services.
Many are curious about the data used to train Sora, which Mira revealed is sourced from public videos from various platforms, as well as videos the company purchased licenses to train. When asked if platforms like YouTube and Facebook were included, Mira declined to comment, only confirming the use of purchased videos from Shutterstock. The training process involves breaking down each object in the videos to align with specific prompt commands.
For the full interview and additional insights, including Sora’s current limitations and strategies to prevent inappropriate video creations, refer to The Wall Street Journal and YouTube for the complete coverage.
TLDR: OpenAI introduced Sora, an AI model for creating Text-to-Video clips, showcasing high detail and up to 1-minute length. The platform is still undergoing risk assessments before a public release and draws data from various public and licensed sources for training purposes. Considerations on cost optimization and energy efficiency are key priorities for the team.
Leave a Comment