作者位次 #2/8ACL 2026Findings· Findings of ACL
Mechanistic Interpretability of Large-Scale Counting in LLM s through a S ystem-2 Strategy
Hosein Hasani · Mohammadali Banayeeanzade · Ali Nafisi · Sadegh Mohammadian · Fatemeh Askari · Mobin Bagherian · Amirmohammad Izadi · Mahdieh Soleymani Baghshah
摘要
Large language models (LLMs), despite strong performance on complex mathematical problems, exhibit systematic limitations in counting tasks. This issue arises from the architectural limits of transformers, where counting is performed across layers, leading to degraded precision for larger countin...