The benchmark problem nobody talks about: if you sell evaluation data to the same labs you're supposed to be measuring, they'll optimize for your metrics instead of real progress. Campbell Brown is describing the structural conflict built into AI evaluation right now. Labs need good datasets to improve their models. But if the entity providing those datasets also sets the benchmarks, you've created an incentive to teach to the test — labs optimize for the score, not actual capability. The soluti

The truth about: Is benchmarking models bullshit? 👇 #lovablepartner #shorts