The AI competition among Silicon Valley giants is becoming increasingly fierce, and they are attacking at full speed.
On October 4th Beijing time, OpenAI and Meta simultaneously unveiled their new weapons.
Meta has released a new AI model called Movie Gen. It consists of two models, Movie Gen Video and Movie Gen Audio, which can seamlessly produce complete multimedia content and achieve comprehensive coverage from images, vision to hearing. Meta claims that this is the "most advanced media infrastructure model to date".
OpenAI has heavily launched the interactive interface canvas, stating that it is a new way of writing and programming using ChatGPT, and a major visual interface upgrade since the release of ChatGPT. Users can collaborate with ChatGPT to complete writing and coding projects, and the entire process is traceable and visible, no longer limited to simple chatting.
It can be foreseen that in the foreseeable future, technology giants will continue to innovate, including revolutionary new products and technologies, as well as sustained new releases that will constantly create small waves.
It is uncertain when this AI competition will be resolved, but what is certain is that the computing power resources and constantly expanding business boundaries of giants at the top will lead to a shrinking survival space for AI startups.
The "Matthew Effect" is gradually becoming clear in the emerging field of AI.
Meta version Sora launched
Calling Meta's Movie Gen "Sora" is not very accurate, but rather "better than blue". In addition to Sora's video generation function, Movie Gen can also generate background music and sound effects for videos, and edit videos directly according to instructions, step by step.
Specifically, Movie Gen Video is a 30B parameter Transformer model that can generate high-definition videos of 16 seconds and 16 frames per second from a single text prompt. The simultaneously released Movie Gen Audio accepts video input and optional text prompts, generating high fidelity audio synchronized with the video.
Zuckerberg, who just became a hot search on the "strongest on the surface" holographic AR glasses platform, has transformed his static photos into various dynamic fitness videos on social media platforms for Movie Gen, with humorous and sci-fi scenes.
The other example videos released by Meta demonstrate the lifelike effects of Movie Gen in generating object motion trajectories, maintaining the physical motion laws of the subject, unifying the background.
As an integrated AI media tool, Movie Gen also has video editing capabilities, which can perform global modifications such as adding, deleting, or replacing elements, replacing backgrounds, and changing styles.
An overseas netizen sighed and said, "With a large number of creators learning to use AI video editing tools, it's hard to imagine what TikTok (short video representative) and YouTube (long video representative) will become in a few years
Similar to Sora's release pace, Movie Gen also needs some time to be made public, but unlike OpenAI's often joked "close", Meta has released a 92 page paper that introduces its architecture, training methods, data management, evaluation, parallel training and inference optimization, as well as audio models.
According to Meta, the video generation model was pre trained on 100 million videos and 1 billion images, while the audio generation model used approximately 1 million hours of audio data.
In addition, Meta has also open-source multiple benchmark datasets, including Movie Gen Video Bench, Movie Gen Edit Bench, and Movie Gen Audio Bench, providing authoritative evaluation tools for future researchers.
Movie Gen has taken a big step towards AI video editing based on OpenAI's presentation of Sora in February this year. Sora, which took the lead, has not yet been officially released. However, Sora can generate videos with a duration of one minute, while Movie Gen currently only has 16 seconds.
Trillion OpenAI's new actions continue
Meta fired a shot at OpenAI, but the latter was not idle.
On September 13th, OpenAI officially released its new universal language model, o1, which is based on slow thinking and outperforms human PhDs in reasoning ability. It also performs exceptionally well in fields such as mathematics, physics, chemistry, and biology.
Its text editing system is also evolving. On October 4th, when Movie Gen was released, OpenAI announced the launch of a visual product called Canvas, which is actually a new human-computer interaction interface.
Canvas is a document and code editor built into ChatGPT. Users no longer need to switch back and forth between ChatGPT and the document editor. When writing a document, it will automatically pop up, allowing them to directly edit any document or code in ChatGPT, completing the entire process from ideation to writing in one interface, achieving seamless collaboration with AI and greatly improving work efficiency.
Industry insiders commented that OpenAI hinted through Canvas that ChatGPT is not just a chatbot, but aims to create an operating system for the AI era. This exposed the extraordinary ambition of its CEO Sam Altman. However, Canvas will only be launched to Enterprise and Edu users next week, while Plus users can directly access it.
The day before, on October 2nd local time, OpenAI announced the completion of a new round of $6.6 billion financing, with a post investment valuation of $157 billion (approximately RMB 1.1 trillion). This marks OpenAI's transition from a "star unicorn" company to a full-fledged artificial intelligence giant.
This round of financing was led by venture capital giant Thrive Capital with $1.3 billion in investment, followed by Microsoft with $750 million, SoftBank Group and Nvidia with $500 million and $100 million respectively, and Apple decided to withdraw at the last minute. In addition, OpenAI has also obtained a new revolving credit line of $4 billion. It is reported that ChatGPT has exceeded 250 million weekly users and is expected to generate revenue of 3.7 billion US dollars this year.
It is reported that OpenAI's funding for this round of financing will be provided in the form of convertible notes, but if it fails to complete the restructuring of its corporate structure within two years (i.e. transforming from a non-profit company to a for-profit company), investors in this round of financing will have the right to demand a refund of their investment.
It is worth noting that OpenAI has also been exposed to require investors participating in financing to choose between two options, namely not to support its competitors' startups, including Anthropic, an AI startup founded by former OpenAI employees, xAI founded by Elon Musk, and SSI, a new company founded by OpenAI co-founder and former chief scientist Ilya Sutskever. News has spread that Musk has once again launched a counterattack against Ultraman on social media.
Along with the transformation of OpenAI's nature and divergent development directions, there has been a sustained major upheaval among the company's executives.
On September 26th, a week ago, OpenAI CTO Mira Murati, Chief Research Officer Bob McGrew, and Vice President of Research Barret Zoph announced their resignations simultaneously. A few days later on October 2nd, OpenAI co-founder Diederik Kingma announced his joining Anthropic. And the company's president, Greg Brockman, is still on a long-term vacation. Among the initial 11 founding members of OpenAI, apart from Ultraman Gong Dou who returned, only Wojciech Zaremba, the head of the language and code generation team, remained.
It seems that 'troubled times' are not enough to summarize the level of turbulence at OpenAI. It is the norm for OpenAI to' ignite 'while' rapidly moving forward '.