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

Source: Dev.to

Introduction

Hello everyone, I’m Or Lenchner, CEO of Bright Data.
Three years after the launch of ChatGPT‑3, the “magic” of large language models (LLMs) continues: they can answer highly complex questions, win gold medals in Math Olympics, and demonstrate astonishing reasoning abilities.

However, despite their intelligence, these models often stumble on everyday tasks that require up‑to‑date, real‑world information—such as purchasing a product or answering a simple factual query. When a model lacks current knowledge, its intelligence alone isn’t enough.

From the Data Perspective

Bright Data’s Role in the AI Ecosystem

Bright Data serves ≈ 20 000 customers who need reliable, large‑scale web data for training and real‑time inference. Our clients include most of the major foundational‑model builders, e‑commerce platforms, financial institutions, and cybersecurity firms. Whenever they require fresh web data—whether for model training, price comparison, investment decisions, or threat detection—they turn to Bright Data.

Scale of Web Data Collection

• ≈ 50 billion web pages processed daily – roughly three times the total number of Google search queries worldwide each day.
• Our web‑archive now exceeds 450 billion pages and continues to grow rapidly.
• We operate the largest known pool of bots that browse the internet continuously to collect this data.

Industry Impact

• E‑commerce: Major platforms rely on our data to stay competitive, adjusting pricing and inventory based on real‑time market conditions.
• Financial services: Large banks use up‑to‑date web data for investment decisions and risk assessment.
• Cybersecurity: Threat‑intelligence teams collect fresh indicators of compromise from the open web.

Merging Intelligence and Knowledge

When an open‑source LLM (e.g., a ChatGPT‑style model) is combined with Bright Data’s MCP (Managed Crawling Platform), it gains access to real‑time web information. In the demo, the model retrieves up‑to‑date salary data, illustrating how knowledge layers can turn a purely intelligent model into a truly useful assistant.

The speaker predicts that by 2026 the distinction between “intelligent” and “knowledgeable” models will blur, enabling full automation of everyday tasks across e‑commerce, travel, and beyond.

This article was auto‑generated from the original presentation. While we strive for accuracy, minor typos or inaccuracies may be present.