AWS re:Invent 2025 - Intelligent vs. Knowledgeable Models through the Lens of Data (AIM358)

Summary of "Intelligent vs. Knowledgeable Models through the Lens of Data" (AWS re:Invent 2025)

Introduction to Intelligent vs. Knowledgeable Models

• The presenter, Orur Lenchner (CEO of Bright Data), discusses the difference between intelligent models and knowledgeable models.
• Intelligent models are extremely sophisticated and can perform complex tasks, but may lack the necessary knowledge to handle simple, everyday tasks.
• Knowledgeable models combine intelligence with real-world knowledge, enabling them to tackle a wider range of practical applications.

Bright Data's Role in the AI Ecosystem

• Bright Data serves over 20,000 customers, including major language model and AI companies, who rely on Bright Data's web data collection capabilities.
• Bright Data processes over 50 billion web pages per day, which is more than three times the global daily Google search volume.
• Bright Data's extensive web data archive and web scraping infrastructure position the company as a leading provider of data for the AI industry.

Trends in AI Model Development

• Open-source and open-weight models are now on par with closed-source, closed-weight models in terms of performance.
• The cost of training large language models is decreasing due to improvements in GPU technology and reduced energy/cooling costs.
• As these barriers are lowered, the availability and quality of data becomes the critical factor for building effective AI models.

The Limitations of Intelligent Models

• Intelligent models, like theoretical physicists, can describe and propose theories, but cannot perform real-world experiments or tasks.
• Using the example of buying milk, the presenter demonstrates that an intelligent model can provide information about the milk, but cannot complete the practical task of purchasing it.

The Emergence of Knowledgeable Models

• The presenter predicts that in 2026, there will be a significant shift towards merging the intelligent and knowledge layers of AI models.
• This will enable models to not only have deep understanding, but also access real-time, relevant information to complete practical tasks.
• Examples include e-commerce and travel, where models can provide information about products and services, as well as real-time availability, pricing, and other key details.

Key Takeaways

• Intelligent models are highly sophisticated but lack the necessary knowledge to perform simple, everyday tasks.
• Knowledgeable models that combine intelligence with real-world knowledge are emerging as the next frontier in AI development.
• The availability and quality of data, facilitated by companies like Bright Data, is becoming the critical factor for building effective AI models.
• The merger of intelligent and knowledge layers in 2026 is expected to unlock significant opportunities for automating a wide range of human tasks.