
Author of "LLMs Can Invent Their Own Compression" in Rajan
How media typically covers Rajan
Rajan as author
LLMs can learn their own compression schemes that achieve near full-context performance while using an order of magnitude fewer tokens, discovering techniques like aggressive pruning, dense punctuation, and code-switching to Mandarin.
“Author of "LLMs Can Invent Their Own Compression" in Rajan”