Automatic or Controlled? Repetition Priming Reveals Divergent Processing in Base LLMs, Instruct LLMs, and Humans
New research reveals how alignment training shifts LLM cognitive processing from raw pattern matching to context focus.
Words recur constantly in natural language use, yet it remains unclear whether language models reactivate prior representations or re-evaluate repeated words afresh, and whether post-training changes this default behavior. We apply repetition priming (Shiffrin and Schneider, 1977...
ArXiv CS.CL
Source
Generated insight by OnAirToday AI
Post-training protocols fundamentally reshape how LLMs handle repeated information, transitioning them from automatic, persistent memory retrieval to context-dependent, controlled processing. This divergence highlights a critical gap between synthetic cognitive architectures and the hybrid, adaptive nature of human linguistic memory.
Comments
Keep comments relevant and respectful. No links, advertising, self-promotion, or spam.