Mechanistic interpretability researchers applying causality theory to LLMs
Why CEREBRO kept it
LLM causality/interpretability research, core mechanistic work
The text below is an automated extraction of the article at https://cacm.acm.org/news/can-we-understand-how-large-language-models-reason/, stored verbatim in the public cerebro-vault repository. Copyright remains with the original publisher (cacm.acm.org).
Mechanistic interpretability researchers applying causality theory to LLMs
Community take
Mechanistic interpretability can't distinguish whether models actually reason or just map inputs to outputs, making the paper's optimism about 'understanding' neural networks overblown.
Backlinks
Appeared in 1 briefing