CURSOR'S INSTANT FULL FILE EDITS WITH SPECULATIVE EDITING

Summary from the TLDR AI newsletter

Researchers used a custom Llama 3 70B model with a speculative prior to edit files at 1,000 tokens per second, giving them near instant rewrites. They did this without diffs and with some clever output formatting.