If metadata loging is a concern, worth mentioning you can self host the co ordinates server with the headscale instead. cuts tailscale is out of loop entirely rather than just opting out of what claim to collect.
Wasn't talking about the client flag, agreed that's seperate headscale replaces the coordination server itself the part is tailscale inc actually runs and sees the connection through that's the server side pice
Worth checking, elsewhere in this thread someone points out glm only assessed the damage after the fact, it didn't stop the attack, and Hf apprently never sought access to a trusted defender program with a closed model either the open model saved them farming might not hold up.
Any one knows how this holds up on long context retrieval (needle in haystack , ruler) vs same size full attention model? efficiency gains look great but that usually where linear attention hybrids fall apart.
Im 19 trying to break into teach and third exactly what scares me they killed the apprenticeship before i could get in how do you become Sara if there's no Ben left to learn from.
