“Edwin Chen believes that many frontier LLMs have been "benchmark hacked," meaning they are narrowly optimized for academic or synthetic benchmarks which do not reflect real-world, open-ended problems.”