Bias on Latent Variablehttps://latentvariable.ai/tags/bias/Recent content in Bias on Latent VariableHugoen-usWed, 15 Apr 2026 07:30:00 +0000Alignment Doesn't Composehttps://latentvariable.ai/posts/alignment-doesnt-compose/Wed, 15 Apr 2026 07:30:00 +0000https://latentvariable.ai/posts/alignment-doesnt-compose/An ICLR 2026 paper proves that individually aligned agents amplify bias when composed into multi-agent systems. The architecture itself is the problem — not the agents. Worse, providing objective context accelerates polarization rather than reducing it.