How Does AI Understand Images and Language?

To a computer, an image is just a table of numbers, and a sentence is just a few numbers marking it. Yet you give it an image and ask it a question, and it can answer. The secret lies in this: people have found a way to transform both images and text into the SAME FORM. This video goes from a tiny pixel to that point. 00:00 - Introduction: How computers "see" an image 01:14 - The video's 4 main parts 02:05 - Part 1: The nature of images from a computer's perspective (Pixels & Number Tables) 03:48 - Semantic distance: The hardest problem in AI 04:12 - Part 2: The history and how computers read text (From 1914 to the present) 05:33 - The combination of CNN, LSTM and the Transformer breakthrough 07:07 - Attention mechanism - The main character of Transformer 08:41 - How to hash text into Tokens and Vectors 10:22 - Part 3: When images are also considered a string of words (Vision Transformer) 12:06 - How images are processed inside the Transformer brain? 13:42 - CLIP: How to Bridge the Worlds of Images and Text 15:16 - Zero-shot: Recognizing Something Never Learned Before 16:47 - Part 4: Assembling a Visual Language Brain (Vision Encoder & Projector) 18:26 - How Images and Text Enter into a Single Sequence 19:46 - Why Does AI Know Where to "Look" to Answer Questions? 21:23 - New Direction: DeepSeek OCR and Information Compression 23:07 - Common Mistakes and Limitations of AI (Illusions, Incorrect Counting...) 24:46 - Summary of the Entire Journey from Pixel to Meaning #AI #MachineLearning #ComputerScience #ComputerVision #HocGiaiThuatCungHPN