arXiv AI
The Instruction Tax on Factual Recall: Base Models Can Beat Their Instruction-Tuned Counterparts
Read the original on arXiv AI →The Flow has not summarised this story yet — read it at arXiv AI.
Related stories
arXiv AI
Sep 3
DKL: Decoupled Knowledge Learning for Instruction-Tuned Language Models
The paper introduces DKL, a method for adding new knowledge to instruction‑tuned language models without compromising their instruction‑following abilities. DKL performs extended pre‑training on a base LLM to embed knowledge, then merges these weights into the instruction‑tuned model, avoiding costly instruction fine‑tuning. Experiments show DKL raises RAG accuracy from 54.17% to 79.26% on retrieval failure cases while using far less training data than previous approaches.
Hugging Face Trending Papers
Sep 2
DKL: Decoupled Knowledge Learning for Instruction-Tuned Language Models
The paper introduces DKL, a method that decouples knowledge learning from instruction tuning in language models. Instead of fine‑tuning the instruction‑tuned model directly, DKL first extends pre‑training on a base model to embed new knowledge, then merges these weights into the instruction‑tuned model, preserving its instruction‑following abilities. Experiments show DKL raises RAG accuracy from 54.17 % to 79.26 % on retrieval failures, outperforming prior methods while using far less training data.
arXiv Computation and Language
2d ago
Memory-Aware Reward Modeling: Learning Human Preferences for LLM Memory Use
The paper investigates whether using a language model’s memory to personalize responses aligns with user preferences. By collecting about 1,000 pairwise judgments from 87 participants on ChatGPT queries that included personal memories, the authors find that memory‑conditioned responses are not consistently preferred and that users sometimes strongly favor non‑memory responses. An automatic reward model trained on these judgments achieves 72.8% preference prediction accuracy and outperforms existing baselines, highlighting the need for evaluation and optimization objectives that better reflect user preferences for memory use.
arXiv AI
Jun 10
Recalling Too Well: Sycophancy Evaluation and Mitigation in Memory-Augmented Models
arXiv:2606. 10949v1 Announce Type: new Abstract: Persistent memory systems promise to make LLMs more helpful by storing user beliefs over time.
arXiv Computation and Language
3d ago
DiaLong: Benchmarking Reminiscence in Long-Context Dialogues
arXiv:2610.25463v2 Announce Type: replace Abstract: How well can transformer-based Large Language Models (LLMs) retain and recall memories in long contexts of dialogue? LLMs are known to forget long-...
arXiv Machine Learning
Jul 14
FineInstructions: Scaling Synthetic Instructions to Pre-Training Scale
arXiv:2601. 22146v2 Announce Type: replace-cross Abstract: Due to limited supervised training data, large language models (LLMs) are typically pre-trained via a self-supervised "predict the next word" objective on a vast amount of unstructured text data.