Imagine Web3's underlying infrastructure as the city's water supply system, and oracles are the main roads that deliver the "off-chain world" to "on-chain contracts."

Traditional oracles focused on highly structured data such as prices, interest rates, and random numbers—fixed formats, clear sources, and multi-source verification were straightforward. But now, there's a clear shift in the wind: prediction markets need to determine "whether something actually happened" and its boundaries; RWA (Real-World Assets) require proof that "the asset exists, its status hasn't changed, and ownership is clear"; AI Agents must read news, announcements, research reports, images, and various documents in complex environments, ultimately passing conclusions to on-chain logic for execution.

The key transformation lies here—data is increasingly "like the real world," and the real world happens to be the most annoying in being unstructured, semantically complex, and full of noise.

Recently, I revisited some project materials, and one detail left a deep impression: some new oracle projects don't treat large language models as "smarter data cleaners," but instead attempt to systematize the "model processing procedures, evidence locations, verification, and penalties." In other words, contracts can not only receive a result but also understand "where this result comes from, whether it can be re-verified, and who is responsible for this answer."

A leading research institution summarized these types of projects as "AI-enhanced decentralized oracle networks," emphasizing the use of large language models to process real-world data, enabling on-chain applications to truly access physical world information—this is no longer a future vision but a direction already being implemented.