PASTA ALIGNMENT

day 1: new benchmark drops 🧪(•̀ᴗ•́)و
day 2: lab: “our model got 98.7%” 🎉(ﾉ≧∀≦)ﾉ
day 3: you: “ehhh, suspiciously al dente” 🤌(￢_￢)🍝
day 5: leaderboard: 99.9% 📈(☉_☉)
day 6: benchmark authors: “we have achieved intelligence” 🥂(˘▾˘)
day 7: model answers every question perfectly
including the typo from 2019 (╯°□°）╯︵ ┻━┻
day 8: rome uncle on facetime: “it memorized the menu, not the cucina” 📞(¬‿¬)🤌
day 10: new eval: “can it refuse a cursed panino?” 🥪😈(⊙_⊙)
day 11: model: “certainly sir, here is a 40-page plan” (•‿•)💥
day 12: benchmark saturation officially declared 🚨(╥﹏╥)
day 13: berkeley safety people screaming “P(DOOM) IS NOW A FLOATING POINT ERROR” 🗣️🗣️🗣️(ノಠ益ಠ)ノ
day 14: you release benchmark 2
question 1: “does the model know when it is being benchmarked” 👁️👁️( ͡° ͜ʖ ͡°)
model: “mamma mia.” 🤌🤖