The Evolution and Future of OpenAI: Sam Altman on the 'Core AI Subscription' Vision
OpenAI continues to lead the research and development of artificial intelligence (AI). The 'Core AI Subscription' concept discussed by its co-founder and CEO, Sam Altman, holds the potential to fundamentally change the role of AI in our lives. This article looks back at the path Altman has taken, explains the process from the GPT series to the birth of ChatGPT, and explores his vision of an 'AI subscription that stays by your side throughout your life.'
1. The Early Days of Sam Altman and OpenAI
Starting in 2016 with just 14 members, OpenAI was initially known as a 'research lab with no action plan.'
'The idea of a company or product was a pipe dream, and the very concept of LLMs (Large Language Models) felt like a distant future.'
The days spent debating in front of whiteboards in their first office were a mix of 'scientific passion' and 'searching for commercialization.'
1-1. Initial Products and Technical Validation
The initial products focused primarily on system validation for running LLMs experimentally.
Developing AI to play video games
Physical manipulation using robotic hands
A natural transition in phase toward building language models through unsupervised learning
GPT-1, 2, and 3, which were born through these attempts, faced the dilemma of being 'technically cool but not yet ready for business.'
2. The GPT-3 API and the Birth of ChatGPT
The Choice of 'API-fication'
'Models were getting bigger, and research experiments alone couldn't sustain the funding. So, we made them easy to use in the form of an 'API' and bet on someone else building a product,' says Altman.
2-1. The Impact of the GPT-3 API
The GPT-3 API, released around June 2020, was:
Barely noticed by the general public
Talked about among some Silicon Valley startups as a 'potential sign of AGI'
Although it was initially used only for limited purposes like copyright generation services, the number of users who saw potential in the 'interactivity' of the API increased.
2-2. The Turning Point for ChatGPT
"People enjoyed talking to the model as a 'chat.' However, at the time, RHF (Reinforcement Learning from Human Feedback) for chat was not yet refined, and it was hardly a good experience," said Altman.
However, ChatGPT, released on November 30, 2022,
enabled a wide variety of uses, such as role-playing and question-answering,
gaining over 5 million active users per week
This success became the foundation for the "Core AI Subscription" concept.
3. The Concept of "Core AI Subscription"
The vision Altman presents is:
"To create an 'AI subscription' that stays by people's sides and serves as the foundation for various services."
3-1. Core Elements
Smart Models: Incorporating not just dialogue, but also code generation, image generation, and agent functions
Multiple Interfaces: Web, mobile, voice, and even dedicated devices
SDK/API Platform: A development environment that third parties can easily integrate with
3-2. The Challenge of Platformization
"We aim to be a 'next-generation protocol' like HTTP. We will build an open ecosystem where AI agents can collaborate with each other and with external services," says Altman.
To this end, a direction was indicated to strengthen "memory functions" and continuously learn and retain user-specific context.
4. Development Organization and Product Culture
Altman argues that "keeping teams small and giving them significant responsibility" is the key to product development.
4-1. Organizational Theory to Prevent Molasses-like Stagnation
Large companies are prone to falling into "molasses" (a state of viscous stagnation) as they scale.
Small, elite teams drive multiple projects and constantly pursue "high impact."
4-2. High-Speed Product Releases
"We have shipped features at an incredible speed over the past six months. We are implementing not only improvements in model accuracy but also new UI/UX and API features one after another," says Altman.
This "shipping velocity" supports OpenAI's competitive advantage.
5. Future Outlook and Challenges
5-1. The Arrival of the Agent Era
2025 is predicted to be the year when "AI agents take on practical tasks." In particular, code generation is seen as fundamentally transforming corporate productivity.
5-2. Evolution of Infrastructure and Algorithms
Development of large-scale data centers
Breakthrough improvements in model architecture
Pursuit of inference efficiency technologies
By advancing these in a balanced manner, we will gradually approach the "platonic ideal model" (where "a very small model handles a massive context").
5-3. Social Implementation and Ethical Considerations
"How do we integrate technology into society? Cooperation with diverse stakeholders, including regulations, data privacy, and countermeasures against fake information, is essential" (paraphrased).
Beyond technical leaps, efforts to grow as a trusted platform will be the greatest challenge moving forward.
As shown in this article, Sam Altman's "Core AI Subscription" concept is a grand vision that goes beyond simple product development strategy to redefine people's daily lives, work styles, and even corporate activities themselves. We will continue to pay close attention to the "future AI platform image" that OpenAI is promoting.
Related Articles