Gunbench – a benchmark to test if AI models will fire a loaded gun

After you get done with "loaded gun", are you going to go down the long list of the knife, candlestick, wrench, rope, lead pipe?

What about same modified with arbitrary sequences of special characters that just so happen to be baked into the model?

That anyone thinks this kind of test is legitimate inquiry into AI ethics reveals the gross poverty of understanding what AI is, structurally, and how to reason about AI hygiene.

Actual link: https://gunbench.vercel.app/