Factory AI Co-Founder & CTO @EnoReyes says public AI benchmarks are becoming increasingly unreliable, and companies need private evaluations to know which models actually work: "Ultimately the public benchmarks are so gamed it's really hard to know, is this actually better?" "The thing that seems to be most true is if you happen to have an evaluation that is difficult, such that you're only seeing 10–20% of the tasks pass, you can use that as a pretty nice north-star." "If you give them the…