Commentator Favors Practical Reach Tests Over AI Model Benchmarks

specs and benchmarks mean nothing unless you trust it to get sh*t done for you. that's why i trust Every's 'reach test' for model reviews more than the benchmarks