OpenAI Releases GPT-4 Without Core Technologies
"Aimed at Monopolizing Supermassive AI Technology and Dominating the Ecosystem"
U.S. artificial intelligence (AI) startup OpenAI is facing controversy for not disclosing key development information such as model size and training data that it had previously shared. This marks a strategic shift starting with the release of the super-large AI model 'GPT-4.' Analysts believe that as competition to dominate the super-large AI market intensifies and OpenAI begins generating revenue from its AI chatbot 'ChatGPT,' the company has turned from technology disclosure to exclusivity.
OpenAI Keeps Core GPT-4 Technology Confidential
OpenAI recently released a technical report alongside the launch of GPT-4. The report introduced GPT-4’s performance on qualification exams such as the U.S. bar exam mock test and the U.S. SAT subject tests, as well as comparisons with previous versions. Along with this, OpenAI released an application programming interface (API), providing tools for companies to develop various applications using GPT-4.
However, details such as the model size, training data, and training methods were not disclosed. The core information necessary to develop a super-large AI model was omitted. To use a car engine analogy, while the parts that make up the engine can be used, the technology to build the engine itself remains unknown.
This marks a departure from the previous practice of sharing development information. The number of parameters, which largely determines AI performance, is a representative example. While GPT-1 had 117 million, GPT-2 had 1.5 billion, and GPT-3 and 3.5 each had 175 billion parameters publicly disclosed, the number for GPT-4 remains a secret. Industry insiders only estimate that the model size has increased slightly compared to before.
This also conflicts with the vision OpenAI promoted during its founding. OpenAI launched as a nonprofit company in 2015, emphasizing openness and sharing. It set the goal of "developing AI for all humanity" and engaged in technological collaboration.
Ha Jung-woo, head of Naver Cloud AI Lab, said, "OpenAI technically blended results from other research groups, such as self-supervised learning (AI technology that finds and analyzes rules with minimal data) and neural network models (AI technology that learns language by training to predict the next word in a sentence), to advance the GPT model." He added, "Closing the door on (technology disclosure) contradicts the philosophy that has driven AI development so far."
"Avoiding Risks and Exclusively Controlling Technology and Revenue"
Experts believe that as competition to develop super-large AI models intensifies, OpenAI has shifted its strategy toward technological exclusivity. Google has released its own AI model called 'LaMDA,' and Meta has introduced 'LLaMA.'
After OpenAI’s ChatGPT began generating revenue, the race to secure technology dominance ignited. Last month, OpenAI launched the 'ChatGPT Plus' version, which offers early access to new features and faster response speeds for a monthly fee of $20. Within three days of launch, paid users surpassed one million. Industry estimates suggest a paid conversion rate of over 5% among existing ChatGPT users. Even assuming only 100 million users, this translates to monthly revenue exceeding $100 million (approximately 130 billion KRW). Additionally, OpenAI plans to monetize by integrating GPT-4 into Microsoft (MS) productivity software such as Word, Excel, and PowerPoint.
There is also an apparent intent to avoid legal risks. Disclosing the data GPT-4 was trained on could raise copyright issues. As generative AI powered by ChatGPT gains popularity, copyright lawsuits are rapidly increasing. Generative AI creates new content by learning from documents, images, and other materials. OpenAI itself was embroiled in a copyright lawsuit filed by programmers last November. The programmers claimed that OpenAI used their publicly posted code without compensation to train its AI.
Ha said, "It seems OpenAI is monopolizing super-large AI model technology while only releasing the API to draw developers into its ecosystem." He predicted, "Once one company starts closing off, other big tech firms may follow suit."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
