Sora’s advanced and controversial artificial intelligence is doomed to failure

Sora’s advanced and controversial artificial intelligence is doomed to failure

OpenAI, the creator of ChatGPT, introduced its latest artificial intelligence model called Sora last week, which can produce realistic videos based on text commands, and since its introduction, there have been many reviews about it.

Yan Likanchief scientist of Meta artificial intelligence, He wrote on Platform X: “Animated modeling of the world by producing pixels is useless and doomed to failure.” Likan believes that the text-to-video conversion model, which is very popular, cannot eventually include all the features.

Likan specifically disagrees with OpenAI’s claims that Sora will eventually enable the creation of generic simulators of the physical world. He argues that Sora’s approach to creating a world simulator is completely wrong.

Likan refers to an old debate in machine learning between generative models and discriminative models. He believes that the first method is very inefficient and cannot adequately deal with the uncertainty caused by complex projections in three-dimensional space.

These models try to infer too many details that are sometimes not even relevant, Likan says. For example, trying to calculate the trajectory of a soccer ball based on how each material it is made of can affect the condition of the ball. This is while the model in such conditions must pay attention to things like mass and speed.

“If your goal is to produce video, there’s no problem; “But if your goal is to understand how the world works, you’re not getting anywhere.”

Likan believes that a generative approach with large language models such as GPT-4 has been the answer so far; because the text is discrete and has a limited number of symbols; But if you want to simulate the world, you’ll need much more than just a few parameters.

Meta, in collaboration with Lecon, are working on their artificial intelligence called V-JEPA, which, unlike current approaches that try to predict and fill in every missing pixel, has the ability to discard unpredictable information, improving training and performance by a factor of 1.5 to 6. leads to

Likan is one of the outspoken godfathers of artificial intelligence, and according to himself, he is not worried about the increasing progress of artificial intelligence, and he is never afraid of criticizing his competitors and will continue his work and path.

Source link

Comments

No comments yet. Why don’t you start the discussion?

Leave a Reply

Your email address will not be published. Required fields are marked *