OpenAI, the creator of ChatGPT, has released Sora, an AI model capable of generating lifelike videos lasting up to 60 seconds from user-provided text inputs. The company’s CEO, Sam Altman, showcased several of these AI-generated videos on his social media handle, formerly known as Twitter.
This began a discussion across various social media platforms regarding the capabilities of the technology. Amidst the debate, Elon Musk, CEO of Tesla and previously affiliated with the Microsoft-backed company, revealed that his electric vehicle company has been producing real-world videos for approximately a year.
Musk replied to a video captioned ‘What does @OpenAI’s Sora have to do with @Tesla’s FSD (Full Self-Driving) v12?’ on X, stating, “Tesla has been capable of real-world video generation with precise physics for approximately a year.”
Elon Musk remarked that the video wasn’t particularly captivating because the training data exclusively stemmed from the cars, resulting in a video that resembles typical Tesla footage, albeit with a dynamically generated (non-recalled) environment.
The video’s host compares the research papers from both companies, illustrating how they approached video generation for distinct purposes, yet ultimately arrived at the same solution.
Lack of training compute to train FSD
He also mentioned that the company currently lacks the necessary computational power for Full Self-Driving (FSD) training. He announced that the company intends to address this issue “later this year.”
Elon Musk stated that they have been constrained by training compute for Full Self-Driving (FSD), thus they haven’t trained with other video data, but it’s certainly feasible. He mentioned they plan to do so later this year when they have some spare capacity.
In a separate post on the platform, Musk shared a lengthy video detailing the operation and training process of Tesla cars’ self-driving mode.
The video sparked a separate discussion, with one person suggesting that Tesla should create a video game.
Musk replied, stating, “I have wanted to do that for a long time :). Our real-world simulation and video generation is the best in the world, but unfortunately making a game can only come after we release unsupervised FSD that is far safer than even supervised FSD.”
