أحلى ورقة بحثية شفتها مؤخرا. حاجة إبداع. مشابهة لفكرة AlphaZero. تدريب الذكاء الاصطناعي بدون أي معلومات من الانسان طبقا للتجارب السابقة بتسمحله أنه يصل لمستويات عالية من الأداء لا يستطيع الوصول لها لو تم "تلويثه" بالقصور الانساني.
Sung Kim@sungkim.bsky.social · 2y agoOne of the hardest part of developing any model is prepping a highly curated dataset, so they propose the Absolute Zero paradigm. Their reasoner learns to both propose tasks that maximize learnability and improve reasoning by solving them, entirely through self-play—with no external data!
![a Windows 3.1 Setup Screen with the background being a gradient from Blue Top to Black Bottom with the text "Windows Setup" written in both Arabic and English and a name prompt window in the foreground asking for the Username (I Entered "Mohannad :]]") and The Company name (I Entered "Vibri's Disk Shop")](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:lfovkllruvnz2rf5tfyaihvl/bafkreid6znrcdns4ltka4ggyy52ibgt3ec7i6la4hqgexl57a5o56i7b6a)
