this tweet is 100% spot on.

we've been thinking a lot about the proper approach to eval models, both for the sake of clarity in the industry, but also for the sake of safe deployment into government + major enterprises.

expect a lot more from us on this front!