OpenAI's Chief Technology Officer, Mira Murati, discussed the company's upcoming text-to-video model, Sora, in an interview with the Wall Street Journal. Murati revealed that Sora is set to launch "definitely this year, but could be a few months", with plans to incorporate sound and eventually embed metadata as watermarks in videos. The technology is currently in early access for some individuals in the film industry for testing purposes. However, Murati's responses raised concerns regarding the transparency of Sora's training data. Specifically, when questioned about the use of YouTube videos for training Sora, Murati admitted uncertainty. This ambiguity has sparked discussions about the ethical implications and reliability of OpenAI's data practices. Additionally, the possibility of releasing Sora before the US election was mentioned, with misinformation and bias being "certainly a consideration" for its timing. Furthermore, the inclusion of content such as nude videos in Sora's capabilities has not been ruled out.
CTO of OpenAI. The vague answer is somewhat concerning… (via WSJ) https://t.co/TB7TFHXAVe
How to know if a company used not publicly available data? Listen to OpenAI CTO https://t.co/jphKqg8XLp
🙍Interviewer: Did you train on YouTube videos? 🙎♀️Mira: I'm actually not sure… Translation: "We have YouTube videos downloaded & available for training; we routinely use them for experiments. I'm not sure if they were used specifically to train the model you're talking about." https://t.co/I86FwCGlxg
OpenAI CTO: We trained on public videos Interviewer: What kind of public videos Mira: Silence Interviewer: Like Youtube Mira: I don't know 🤣🤣🤣 ☠️☠️☠️ https://t.co/gbZF6kvqtx
Mira: We trained on public videos Interviewer: What kind of public videos Mira: Silence Interviewer: Like Youtube Mira: I don't know 🤣🤣🤣 https://t.co/gbZF6kvqtx
Watch: In an interview with @JoannaStern, OpenAI CTO says the company’s text-to-video model will be available this year but ducks questions on how the model was trained https://t.co/uym4vnZZmM https://t.co/uym4vnZZmM
Only one of these can be true: -The CTO of OpenAI doesn't know what Sora was trained on -She does, but blatantly lies during softball interviews This is based on what **she said** on video. Question: Which of these options helps you trust this company with advanced AGI?
Mira Murati says she's "not sure" if Sora was trained on YouTube videos 😬😬😬 Google attorneys currently giggling https://t.co/M3KGx25JSQ
Hi! I'd like to talk to you about a company OpenAI. What does it do? Good question. Maybe u can help us find an answer. What if it is closed? We won’t tell you what to do with the weights, we will hide your weights stat. Very interested, somewhat interested, not interested?… https://t.co/YvmJBzOkMa
Hi! I'd like to talk to you about a company OpenAI. What does it do? Good question. Maybe u can help us find an answer. What if it is closed? We won’t tell you what to do with the weights, we will hide your weights stat. very interested, somewhat interested or not… https://t.co/o8yHdoLd07
“OpenAI Says Sora Will Launch in 2024 and Nude Videos Aren’t Off the Table” SoraPorn. Who’s your pimp? Article: https://t.co/oUIylQqDQU
"What data was used to train Sora...so videos from YouTube"? CTO of OpenAI, "I'm actually not sure about that." CEO of OpenAI: Just give us $7T and we'll save the world. Politicians & Press: Write about them with press releases and propaganda masquerading as laws & journalism.
Insightful @WSJ interview of CTO Mira Murati by @JoannaStern on the successes/failures, training, timelines and launch concerns around #Sora video generation. @MattVidPro @mreflow @icreatelife https://t.co/cZ0l2NZuoN
openai will release sora “definitely this year, but could be a few months” mira doesn’t rule out releasing before the US election, it’s “certainly a consideration” — if they’re confident about misinformation and bias https://t.co/tS7OxiCo2S
Mira Murati talked to the WSJ about Sora. Sora will be available this year. Plans are in place to add sound, and eventually, Sora videos will contain metadata as a watermark. OpenAI is currently giving some individuals in the film industry early access for testing. https://t.co/wCNmo4d62u