Are AI models getting safer over time, or is it just bullshit? | Not Hacker News!