I IMPROVED 15 LLMS AT CODING IN ONE AFTERNOON. ONLY THE HARNESS CHANGED

Summary from the TLDR AI newsletter

There's no real consensus on the best solution to the simple 'how do you change things' problem. None of the current tools give models a stable, verifiable identifier for the lines to be changed without wasting tremendous amounts of context and depending on perfect recall. This post discusses a solution where, when a model reads a file, or greps for something, every line comes back tagged with a 2-3 character content hash, and when the model edits, it references those tags. This approach can result in an over 8% improvement in Gemini's success rate with zero training compute cost.