Context Window (குறுகிய கால நினைவாற்றல் / Short-Term Memory Limit)

Let's look at the Goldfish's memory! (தங்க மீன் ஞாபக சக்திய பாப்போமா!)

Technical Meaning: சூழல் சாளரம் (Soozhal Saalaram) - The maximum amount of text an AI model can remember and process at one single time.

The Core Idea

Every LLM has a strict memory limit called a Context Window, measured in tokens. If a model has a 4,000 token context window (about 3,000 words), it completely forgets anything you said earlier in the conversation once you exceed that limit. It is the short-term working memory of the AI.

Expanding the context window is highly expensive. When Google announced Gemini 1.5 Pro with a 1-million-token context window, it was a massive engineering breakthrough.

The Origin Story

Early models like GPT-2 could only remember a few paragraphs. This meant long conversations were impossible—the AI would contradict itself because it forgot what it said 10 minutes ago. As hardware improved, engineers figured out ways to optimize the "Self-Attention" mechanism (which usually requires exponentially more RAM as text gets longer) to support massive books and entire codebases in a single prompt.

The Tamil Analogy

Imagine a person with Short-Term Memory Loss (like Suriya in Ghajini).

He can only remember what happened in the last 15 minutes. That 15 minutes is his Context Window.

If you tell him your name at 9:00 AM, and then talk to him until 9:20 AM, his 15-minute window has shifted. The memory of you telling your name has fallen out of the window. If you ask him your name at 9:21 AM, he will have absolutely no idea!

To fix this, modern AIs (like Gemini) now have massive context windows, allowing them to remember entire movies without forgetting the beginning!

Sources & Further Reading

• Research Paper: Longformer: The Long-Document Transformer (Beltagy et al., 2020)
• Announcement: Google Gemini 1.5 Pro: 1 Million Context Window

Try It Yourself