How far can bias go? -- Tracing bias from pretraining data to alignment

Re-assign community