AI News

LLMs fail N‑back because readout is corrupted by interfering items, not because attention mislocates targets

ABC2 min read
Research