A method for steering llms to better follow instructions - View it on GitHub