arXiv:2606. 20632v2 Announce Type: replace-cross Abstract: Multi-LLM systems use multiple language models to deliberate, judge each other's outputs, or coordinate as agents.
arXiv:2608. 13921v1 Announce Type: new Abstract: LLM agents increasingly maintain personal memory across sessions, but it can conflict.
LLM agents increasingly maintain personal memory across sessions, but it can conflict. Preferences depend on context, behavior evolves, and sources can conflict. When a query lacks context, time, or s...
arXiv:2609.36569v1 Announce Type: cross Abstract: Checkpoint selection is a routine decision in supervised fine-tuning (SFT): training produces multiple checkpoints, but only one is retained. Yet fix...
The study investigates whether post‑training methods—GRPO, SFT, and DPO—improve language models’ ability to follow prompt evidence that conflicts with memorized knowledge. By comparing nine training variants across different scales and families, the authors find that grounding gains are modest for GRPO, moderate for Conflict‑SFT, and near‑ceiling for DPO, but all largely rely on the same causal attention‑head set present in the starting checkpoint. Removing the starting‑model grounding direction suppresses these gains, while adding it back recovers a significant portion of DPO’s improvement, indicating that existing model machinery drives most of the observed gains.
arXiv:2609.38485v1 Announce Type: new Abstract: Unified multimodal models (UMMs) train image understanding and autoregressive image generation on shared parameters, and the two objectives are known t...
arXiv:2606. 24589v1 Announce Type: new Abstract: Scaling adversarial evaluation of large language models requires both a method for generating hard inputs and a reliable way to confirm that resulting failures are real.
The paper investigates how a probe can decode in‑context bindings on model errors and how probe‑guided steering can repair them. It tracks probe accuracy, model output, and steering response across public pretraining and post‑training checkpoints, noting that probe accuracy improves during Pythia pretraining and that steering benefits grow with model size. The study also shows that decoders trained on final state or candidate logits do not outperform each other on late‑checkpoint errors, and presents an information‑theoretic counterexample explaining why decodability on errors alone cannot prove discarded output information.
arXiv:2608.28648v1 Announce Type: new Abstract: We study how instruction-tuned LLMs arbitrate direct conflicts between system and user instructions. We introduce a benchmark of 41 paired constraints...
The paper investigates in-context binding errors in language models, showing that a linear probe can recover correct entity bindings from frozen hidden states even when the model outputs incorrect bindings. Across 16 checkpoints, the probe’s accuracy on failure cases surpasses a baseline by about 0.196, and a probe‑based score improves failure detection over the model’s confidence by 0.079 AUROC. Steering the residual stream toward the probe‑decoded binding further boosts accuracy by an average of 0.168 across eight models.
arXiv:2609.08618v1 Announce Type: new Abstract: Benchmark scores describe what a checkpoint can do now, but they do not determine how it will respond to the next training episode. We measure this mis...
arXiv:2608. 04509v1 Announce Type: new Abstract: Vision-language systems combine images with retrieved text, but these sources can disagree or jointly fail to support an answer.