其他服務業Business News

OpenAI, the maker of ChatGPT, has unveiled its new tool: Sora, an AI text-to-video generator that creates short videos (up to 60 seconds) based on written prompts and from existing still images. Sora is not the first of its kind. There have been other image generators such as DALL-E and Midjourney, however getting an AI model to generate videos is much newer and complex. Google, Meta, Stable Video Diffusion, and Runway ML (a startup) are some other companies who have demonstrated similar technology to OpenAI. However, analysts have noted the high quality of the tool’s videos that have been generated so far and acknowledge that it demonstrates an impressive achievement. The head of U.S. AI and software research at Macquarie, Fred Havemeyer, has stated that the launch of Sora “marks a big step forward for the industry” (Grantham-Philips & O’Brien, 2024). The videos generated by Sora look more normal and “seem to actually respect physics and the real world more” according to Havemeyer (Grantham-Philips & O’Brien, 2024). Compared to Stable Video Diffusion, Sora’s generated videos are more cohesive and consistent throughout. Sora presents a new opportunity to be incorporated into traditional video content and could eventually generate full-blown, narrative videos from a few prompts. 

Although Sora’s capabilities are astonishing, the rapid development in the AI space raises concerns and fears about the potential ethical and societal implications. However, Havemeyer expressed potential risks for 2024 with the election cycle being held in November. Sora could be used to generate videos that may look and sound realistic, which could lead to a number of “issues within politics and beyond … pointing to fraud, propaganda, and misinformation concerns” (Grantham-Philips & O’Brien, 2024). Currently, Sora is not widely available yet as OpenAI is currently discussing with policymakers and artists before officially releasing Sora to the public. 

The company wants to ensure that the proper safety steps are in place before officially releasing Sora. According to OpenAI, they are working with “red teamers — domain experts in areas like misinformation, hateful content, and bias — who will be adversarially testing the model” (Grantham-Philips & O’Brien, 2024). Furthermore, the company is working on developing tools that would help detect misleading and false content, such as a detection classifier that could tell when a video was generated by Sora. Anna Makanju, OpenAI’s Vice President of Global Affairs, pledged to “voluntarily work together” with other technology companies to “combat AI-generated election deepfakes” at the Munich Security Conference (Grantham-Philips & O’Brien, 2024).

Source: https://apnews.com/article/sora-openai-artifical-intelligence-videos-d217e4dab4bd9e3f60b4cfe672b97563