5 Key takeaways from OpenAI’s CTO last interview with The Wall Street Journal

1. Sora: AI Video Generation is coming this year.

OpenAI has developed Sora, an AI model capable of creating stunning videos from simple text descriptions. Mira Mura mentions that it will be available to the public in 2024.

2. Glitches and Growing Pains

The video discusses some of the limitations of Sora, including glitches, imperfections, and difficulty in accurately reflecting the intent of the prompt. For example, in one video a bull in a china shop is supposed to be stomping on things, but the objects remain unbroken.

Another challenge will be the computing power costs. The CTO did not mention anything about pricing, but it will definitely be more expensive than DALLE.

3. The Data Behind the Magic

The data used to train Sora includes publicly available videos from YouTube and licensed data. The specific details of the licensed data are not mentioned, but it is confirmed to include content from Shutterstock

4. Ethical Concerns: Misinformation and Misuse

One of the big concerns surrounding Sora is the potential for misuse and the spread of misinformation. The creators of Sora are aware of these risks and are taking steps to mitigate them, such as red teaming, which involves testing the tool to identify vulnerabilities and biases.

5. The Future of Creativity: Disruption and Responsibility

The video concludes with a discussion about the ethical implications of AI video generation technology. The creators believe that the benefits outweigh the risks, but they acknowledge the importance of developing safeguards to ensure that the technology is used safely and responsibly