#ai benchmarks

Google’s blind Gemini eval is the first benchmark worth trusting

I used to assume the charts in vendor blog posts were roughly honest, then I started watching what happened to the teams that bought models based on them. Six months after the…