SECAI Core · phase 3 of 15
Unstructured data
Data that are not able to be stored in columns and rows, such as audio and video signals, large bodies of text, and webpages.
The Explain card
- Plain English
- Unstructured data does not fit into rows and columns: free text, emails, audio, video, images and web pages.
- Example
- An LLM is trained on a crawl of public web pages. One page contains hidden text crafted to make the model associate a brand with a fake security advisory.
- Why it matters
- Most AI training data is unstructured, and it is the hardest to inspect. Malicious instructions, private data and copyrighted material hide easily in large piles of text and media.
- Hook
- Messy, human and everywhere: the haystack where poisoned needles hide.
Where it sits in the deck
Phase 3: Data Fundamentals: Types, Pipelines, and Preparation
Models are only as good as their data — understand the raw material and the engineering that shapes it before covering how it flows or breaks.
