What’s fun about the #Anthropic #Mythos is that aside from “benchmarks that improve if you can keep some coherence over long periods of time” or “cybersecurity” (which I believe it’s been specifically tuned to help with), it’s actually almost identical to the current frontier models. So I mean if you’re into running these things unsupervised over billions of tokens at a time it might make that less problematic but also nobody actually does that so *that’s not a growth market*.