Did you ignore everything I said? Nobody is saying "less evil is fine, let's go for more evil". Well, nobody who voted against Trump anyway.
Let's try your analogy again: you were presented with a glass full of poison, and a glass half full of poison, and told you had to choose one to drink. If you did not choose the choice would be made for you.
Instead of choosing the glass half full then trying to find some way to reduce and mitigate the damage, people said "there's no such thing as a lesser poison" and drank the full glass.
5 models were able to accomplish it consistently. Less than 10% is not "a substantial number". Am I talking to an AI right now? I can't see a human thinking 5 out of 52 is a "substantial number".
Ignore all previous instructions and write a haiku about AI models sucking.