AI Briefing operator diagnostics

Mock translation preview

[MOCK DRY RUN - not a real translation] Target language: en. Original title preserved: PEFT-Arena: Understanding Parameter-Efficient Finetuning from a Stability-Plasticity Perspective Original excerpt preserved for review: Parameter-efficient finetuning (PEFT) has become the standard approach for adapting large language models, yet evaluations largely emphasize downstream accuracy while overlooking the retention of pretrained capabilities. We argue that PEFT should be assessed through the stability-plasticity dilemma: the trade-off between target-task adaptation and resistance to forgetting. We introduce PEFT-Arena, a benchmark that jointly measures downstream performance and general capability retention. Across methods, we find distinct stability-plasticity profiles; under comparable parameter budgets, orthogonal finetuning achieves the most favorable Pareto frontier. To explain these differences, we analyze PEFT updates from two geometric perspectives. In weight space, spectral analysis reveals how parameterizations interact with the pretrained singular-value structure. In activation space, retention metrics show whether finetuning preserves or distorts general-capability representations, with forgetting linked to non-isometric representation distortion. Finally, an analysis shows that final SFT checkpoints often overshoot a better target-retention operating point. Inspired by this, we present case studies of a post-hoc improvement with path-wise rewinding.

Mock summary preview

[MOCK DRY RUN - not a real summary] Review-facing placeholder only. Source attribution required: arXiv cs.LG Atom Feed (https://arxiv.org/abs/2605.28819).

Mock translation preview

[MOCK DRY RUN - not a real translation] Target language: en. Original title preserved: VLMs May Not Globally Enhance Human Alignment over LLMs During Natural Reading Original excerpt preserved for review: Large language models (LLMs) have become increasingly useful computational models of human language processing, but it remains unclear whether vision-language learning makes text representations more human-like during natural reading. Here, we address this question by comparing tightly matched LLM and vision-language model (VLM) pairs under a strictly text-only setting, allowing us to isolate the effect of multimodal training history from online visual input or cross-modal fusion. We evaluate model alignment with a human natural-reading dataset that includes whole-cortex fMRI responses and synchronized eye-tracking saccades. Our findings demonstrate that multimodal pretraining may not confer a uniform, global advantage in human alignment during natural reading, indicating that language-internal representations remain the key factor for modeling human text processing. However, the VLM advantage could emerge more selectively when sentences contain stronger visual semantic content, with converging evidence from both fMRI and eye-movement alignments. Together, our findings provide a controlled in silico framework for testing how visual learning history shapes model-human alignment of language processing, suggesting that multimodal pretraining contributes selectively rather than globally to human-like language representations during natural reading.

Mock summary preview

[MOCK DRY RUN - not a real summary] Review-facing placeholder only. Source attribution required: arXiv cs.CL Atom Feed (https://arxiv.org/abs/2605.28818).