@krishnachaytanya: A context window is the limited amount of text an AI model can read at once, and it's why ChatGPT and Claude forget what you told them earlier in a long chat. Every message you send fills the window. When it's full, the oldest messages scroll off the top and the model literally can't read them anymore, so it starts guessing. The fix: save the important instruction and feed it back into the window (retrieval augmented generation, RAG) so it stays in view. GPT-4o holds 128K tokens, Claude 200K, Gemini 1M, Llama 128K. Bigger window, same rule: full means the oldest falls off. Ever had an AI forget an instruction mid-conversation? Save this for next time it happens. #systemdesign #ai #llm #contextwindow #aiengineering

1719

There are no more comments for this video.

To see more videos from user @krishnachaytanya, please go to the Tikwm homepage.