I'm not really in the AI community, but after watching and learning about the Hugging Face incident, I can't help but wonder if this might've reduced its likelihood!:# Add tasks for which giving up is the correct response!Even if people still accidentally add in impossible tasks that only reward cheating (such as hacking), their effect on the models could be diluted by having added in other tasks that are intentionally impossible for which the correct answer is giving up!They could also add in tasks that make it *appear* that there's a break in the sandbox, be monitoring the agents trying to hack, and intensively negatively reward trying to cheat by hacking!#AI #aisafety
Related
📰 Daily Bulletins - 10/11/2026📉 Collapse : 🟠 4/5🌍 Planet : 🟠 4/5🛡️ War in Europe : 🟠 4/5🤖 AGI risk : 🟠 4/5🔗 https://etik...
📰 Daily Bulletins - 10/11/2026📉 Collapse : 🟠 4/5🌍 Planet : 🟠 4/5🛡️ War in Europe : 🟠 4/5🤖 AGI risk : 🟠 4/5🔗 https://etik.soulignac.eu/bulletin/mardi?lang=en#collapse #climate #war ...
Companies such as Google used to be our portals to the rest of the web. Can the internet as we know it survive when A.I....
Companies such as Google used to be our portals to the rest of the web. Can the internet as we know it survive when A.I. answers all of our questions? #culture #search #aihttps://w...
The authors show that transformer models trained with discrete diffusion and decoded by guided beam search can recover t...
The authors show that transformer models trained with discrete diffusion and decoded by guided beam search can recover the prime factors of semiprime integers directly from data.So...