Comment on Australia tells AI chatbot companies to detail child protection steps
KeenFlame@feddit.nu 11 hours ago
Aight i can spoiler alert the steps
Step 1 fiddle with the system prompt
When that backfires (that’s what Elon did with grok)
Step 2 put in expensive guardrails
When that fails (latent space is inherently untethered)
Step 3 lie