Study kehta hai ki LLMs language proficiency mein achhe hain aur instructions follow kar sakte hain, lekin woh naye skills ko khud se seekhne ki ability nahi rakhte.
**In Short:**
Ek nayi study kehti hai ki large language models (LLMs) jaise ChatGPT humanity ke liye existential threat nahi hain.
– Study kehta hai ki LLMs khud se naye skills nahi seekh sakte aur bina explicit instruction ke kaam nahi kar sakte, isliye ye models control aur predict karne mein aasan hain.
– Research ne yeh bhi dikhaya ki LLMs ki capabilities in-context learning (ICL) par depend karti hain, jahan ye models diye gaye examples ke basis par kaam karte hain.
– Study se yeh bhi pata chala hai ki LLMs ko complex tasks ke liye explicit instructions aur examples dene zaroori hain, warna advanced reasoning mein galtiyan ho sakti hain.
Study ne yeh bhi highlight kiya ki AI ke misuse, jaise fake news banane ya fraud mein use karna, ek real risk hai. Researchers ne kaha ki existential threats par regulations lagane se better hai ki actual risks par focus kiya jaye.
