Physically-isolated Experts Routing Network (PiERN), an architecture that directs computation and reasoning at token level, thereby enabling iterative alternation within a single chain of thought, is proposed, offering an efficient, interpretable, and scalable paradigm for interfacing language models with scientific sy...
Web agents complete user requests by reading and acting on pages that third parties write, so an instruction planted on a page can redirect the agent away from the user's goal. The agent cannot simply ignore the page, because the page also holds the values and controls the task requires. Current defenses fine-tune the...
Sarim Hashmi, Mukul Ranjan, Kshitij Mishra et al.· 0 citations
H-CDLMs is introduced, a simple framework that further improves continuous DLMs with minimal compute and parameter overhead and applies H-CDLM to the flow matching model FLM, obtaining consistent gains with H-FLM and demonstrating that the framework generalizes across continuous generative paradigms.
Knowledge that a language model appears to forget during finetuning often remains stored and can be recovered, a phenomenon called spurious forgetting. Finetuning on new facts can even produce forgetting that undoes itself: recall of the old facts collapses, recovers as training continues on new facts alone, and only t...
V. Palit, Florent Draye, Nicolas Zucchet et al.· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
LLM-based agents solve complex multi-step tasks, but sequential execution incurs substantial latency. In principle, parallelizing work across multiple agents should yield near-linear speedups. Yet existing parallel multi-agent systems often run slower than a single-agent baseline. We attribute this gap to two hidden co...
Yexiong Lin, Shanshan Ye, Yu Yao et al.· 0 citations
Streaming video-language models must decide not only what to answer, but whether the evidence needed for the current question has arrived. Existing systems learn that decision as a separate trigger; we ask whether an unmodified model already computes it. We show that frozen VideoLLMs carry a linearly readable evidence-...
Activation steering has gained popularity as a lightweight inference-time debiasing technique for large language models. However, prior work reports that steering vectors generalise poorly, with unintended effects on model performance and limited transfer to new datasets. Our work analyses what the debiasing direction...
Stephanie Buttigieg, Maeve Madigan, P. Kamalaruban et al.· 1 citation· ⚡1
Whether alignment gets easier or harder as models grow is often argued from isolated findings, as if alignment were one property. We treat it as a family of measurable scaling relations: for each risk category r, the alignment burden needed to hold a fixed safety target is modeled as B_r(N)=a_rN^alpha_r, with N a capab...
This work introduces UNifying REtrieval And Long-Context with a Single Model (UNREAL), a model-native evidence selection framework to span corpus retrieval and long-context inference and establishes model-internal evidence selection as a common foundation for corpus retrieval and evidence-sparse long-context inference.
Edan Kinderman, Elad Hoffer, Y. Blau et al.· 0 citations
Retrieval-augmented generation (RAG) is a widely used approach for grounding large language models (LLMs) in external knowledge. However, configuring a pipeline is an expensive hyperparameter optimization problem over many interacting choices, from chunking and embedding model to reranking and generation. Existing opti...
Lasse B. Strand, Robert Jakob, Kevin O'Sullivan et al.· 0 citations
Large language models routinely answer questions that cannot be answered from the information given, and in dialogue they answer before enough has been said. Unanswerability is linearly decodable from hidden states, but it is unclear which of its forms share a representation and whether the signal is useful in dialogue...
Jerzy Kamiński, Ilya Galyukshev, Artem Kuznetsov et al.· 0 citations
Continual learning (CL) is essential for Large Language Models (LLMs) to sequentially adapt to evolving tasks. To mitigate catastrophic forgetting, recent advances implement low-rank adaptation with orthogonal projections (e.g., O-LoRA) to isolate task parameters. However, we reveal that such strict geometric constrain...
Mao-Qi Liu, Quan Fang, Yu-Fei He· Proceedings of the 2026 Conf...· 1 citation
Flint-interactive gives AI agents a language for creating interactive visualizations. Click a chart, filter data, highlight related content, and explore beyond the view the author originally chose. The post Flint-interactive: A language for creating interactive data visualizations appeared first on Microsoft Research.
