Very much an uphill battle for OpenAI over the coming weeks. Imo it's 6.1 Astra tomorrow or bust - Fable 5.5 is on its way and coming sooner than most people expect, and considering the 5.x -> 5.5 leaps we've been observing across smaller classes, it's going to be a beast. *Haiku* 5.5 is likely to beat 6 Sol. Though Anthropic's biggest weakness is still token efficiency, where OpenAI continue to have the advantage In any case, I would expect a successor to 6 Astra next month at the latest. But will further RL be enough or does their next pretrain need to be wheeled out?

very excited to see this paper get recognized, I believe the technique here is widely applicable. This was joint work done with the SkyDiscover team (@shagarw21, @mertcemri, @shulynnliu), along with our advisors (@istoica05, @matei_zaharia, Mohsen Lesani, and Sylvia Ratnasamy),

very clearly define the minimum bar that you will accept for yourself the smallest amount of workout that you will still count as exercise for the day the one thing at work that if you accomplish today you can claim you moved the mission forward the minimum effort that you need to put into an important relationship in your life then hold yourself to those each day and nothing more, telling yourself you did enough