@GaryMarcus: Am old enough to remember when @GeoffreyHinton told me I was stupid for saying that LLMs regurgitate training data. He …
Summary
Gary Marcus highlights recent DeepMind research confirming that LLMs frequently memorize and regurgitate training data, countering past criticism from Geoffrey Hinton. The post underscores ongoing debates about LLM limitations and their real-world capabilities.
Similar Articles
@geoffreyhinton: I believe you said that they JUST (my caps) regurgitate training data. That IS stupid. Here is a quote from you: "It gl…
Geoffrey Hinton counters Gary Marcus's claim that language models merely regurgitate training data, citing Marcus's own words.
LLMs Can Leak Training Data But Do They Want To? A Propensity-Aware Evaluation of Memorization in LLMs
PropMe is a propensity-aware framework for evaluating LLM memorization, distinguishing between forced reproduction capabilities and natural propensity using SimpleTrace for deterministic attribution across open models and datasets.
@rohanpaul_ai: New Illinois+ Tsinghua University and other labs study finds that LLM agents still have unreliable memory and that it c…
A study by University of Illinois and Tsinghua University finds that LLM agents' memory becomes unreliable when continuously rewritten, with performance dropping from 100% to 54% on ARC-AGI tasks. The paper proposes preserving raw episodes instead of always summarizing them.
Useful memories become faulty when continuously updated by LLMs (30 minute read)
This research demonstrates that continuously updating LLM agent memories through distillation and consolidation loops causes performance regression, even when trained on ground-truth solutions. The study finds that episodic-only retention outperforms text-based consolidation, highlighting significant flaws in current self-improvement paradigms.
STALE: Can LLM Agents Know When Their Memories Are No Longer Valid?
This paper identifies a critical failure mode in LLM agents where they fail to update personalized memories when new evidence conflicts with prior beliefs. It introduces the STALE benchmark and a three-dimensional probing framework, revealing that even the best models achieve only 55.2% accuracy, and proposes CUPMem as a prototype for robust memory revision.