Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper investigates the mechanism behind the shared working-memory signature observed in LLMs and humans on the N-back task, specifically differentiating between precision and interference accounts. By analyzing attention patterns and probe decodings across various LLMs and minimal transformers, the study finds that attention heads consistently target the correct item, even when errors occur. The key result indicates that the working-memory bottleneck in LLMs stems from interference during readout rather than an inability to precisely locate the target item.
LLMs' working memory failures aren't due to fuzzy attention, but rather interference corrupting correct retrievals.
Large language models (LLMs) reproduce a human working-memory signature on the N-back task: performance declines as memory load N increases, and errors concentrate on lures, items that repeat the current stimulus at a neighboring but incorrect serial position. We ask whether this shared behavioral signature reflects a shared mechanism. The literature offers two candidate accounts. On a precision account, softmax attention cannot concentrate its weight on the item N positions back, so adjacent items are retrieved instead. On an interference account, attention selects the correct item, but the content of competing items corrupts the decision. We dissociate these accounts in 15 pretrained LLMs from three families (70 million to 14 billion parameters) and in minimal transformers trained from scratch. In all three families, the attention heads that carry the task attend to the target on most missed matches. Linear probes decode near-perfectly whether a matching letter occurred at the target or at a lure position, even on trials where the model false-alarms to the lure, although its own decisions barely distinguish the two. Sharpening attention yields negligible gains, whereas suppressing the letter identity of non-target items improves sensitivity (d') in nearly all model-by-load conditions. In larger models the task is spread over more heads, and removing interference helps more. In minimal transformers, precision limits a single attention layer, but a second layer removes this constraint. The working-memory limit of LLMs therefore lies in reading items out under interference rather than in locating them, consistent with interference accounts of human working memory.