AI "Input": Context Window, Cloud (Document) Storage, and the Analogy to Power Generation, Transmission, and Distribution

As AI technology, particularly Foundation models, continues to rapidly advance, the amount of input text that these models can process at once is increasing significantly. This capacity is known as the Context Window. Currently, models like ChatGPT can handle 168K tokens at a time, which equates to hundreds of pages of text. However, this is just the beginning, and the size of this Context Window is expected to increase more than tenfold in the future.

What does an expanded Context Window mean from a technical standpoint?

The expansion of the Context Window will bring about significant changes in how AI accesses and utilizes knowledge. Previously, it was necessary to pre-train AI on specific domains or fine-tune models to create domain-specific solutions. However, this process will be greatly reduced as models become capable of processing more information at once, allowing them to accurately respond to comprehensive, real-time information from specific domains. From an engineering perspective, this also means the ability to perform multiple tasks in parallel, which in turn increases the importance of output tokens to handle the results of these parallel processes.

This evolution will significantly enhance AI's ability to respond to the latest and most specialized information. For instance, even if AI has not been pre-trained on new regulations or technological changes in a specific industry, it will be able to learn and respond in real-time by simply inputting the relevant documents or data. This could theoretically allow for the deployment of a chatbot or knowledge system specialized in HR policies within an hour.

What becomes important as the Context Window expands?

To effectively leverage these changes, it will be crucial to focus on how to efficiently collect and pre-process the information fed into AI models. If we think of AI usage as analogous to electricity, Foundation models generate the power, cloud service providers transmit it, and the distribution must be managed at the local level—by the end-users. This is where cloud storage services play a key role. If AI models can store and process data close to the cloud storage, they can transmit data quickly and securely without relying on public internet infrastructure. Services like Google Drive and OneDrive are likely to be increasingly utilized for this purpose, especially since they can be closely integrated with models like Google Gemini and OpenAI GPT.

Looking at the technological outlook, the expansion of the Context Window and the resulting improvement in information accessibility are expected to accelerate AI adoption across various industries. AI solution providers have the opportunity to develop integrated solutions between cloud storage and AI models, offering easy-to-use pre-processing tools and data management systems for businesses, creating new business opportunities. There will be a growing need for strategic consulting on knowledge management systems and system integration, as well as partnerships with cloud storage services to offer more efficient data processing and security solutions.