arXiv:2605.27378v1 Announce Type: new Abstract: Dental image analysis plays a pivotal role in supporting accurate diagnosis and treatment planning in oral healthcare. Although recent advances have produced dental AI models for specific tasks and individual imaging modalities, their isolated designs limit practical use in real-world clinical workflows. In this paper, we present OralAgent, the first dental-specialized AI agent that unifies multimodal reasoning, tool-based decision-making, and knowledge-grounded retrieval within an end-to-end automated framework. It integrates 22 visual analysis tools and 368 widely-used classical dental textbooks, enabling autonomous reasoning, planning, tool use, knowledge retrieval, and multi-step workflow execution. Furthermore, we introduce OralCorpus, a large-scale, high-quality bilingual textual resource containing 134.8M tokens curated for dental retrieval-augmented generation (RAG). To evaluate models' multidisciplinary dental knowledge, we construct OralQA-ZH, a Chinese multiple-choice question benchmark consisting of 798 items across eleven oral subspecialties. Extensive experiments demonstrate that OralAgent achieves state-of-the-art performance on the MMOral-Uni, MMOral-OPG, and OralQA-ZH benchmarks, highlighting its effectiveness, interpretability, and adaptability in real-world clinical settings. The code and models are publicly available at https://github.com/isjinghao/OralAgent.

arXiv:2605.27465v1 Announce Type: new Abstract: The quadratic cost of self-attention in Vision Transformers (ViTs) constitutes a fundamental bottleneck for practical deployment, motivating a vibrant line of research on token reduction. Among existing approaches, token merging (ToMe) has emerged as an elegant training-free solution; yet its design rests on an unspoken premise of token equality, which contravenes the well-documented non-uniformity of self-attention and leads to information loss in high-salience tokens under aggressive compression. We address this limitation with AdaMerge, a token-merging framework based on two complementary mechanisms. First, salience-weighted similarity leverages column-wise feature-affinity centrality as a token-importance proxy and incorporates the resulting salience scores into the bipartite matching score, ensuring that pivotal tokens contribute more strongly to the merged representation. Second, adaptive merging intensity uses pre-computed layer-wise similarity statistics to dynamically modulate the per-layer reduction count in accordance with input-specific redundancy. On ImageNet-1k with ViT-B/16, AdaMerge consistently outperforms ToMe, PiToMe, and DSM across all FLOPs-matched regimes. The accuracy gap widens monotonically with compression: at the 13.4G FLOPs operating point, AdaMerge sustains a Top-1 degradation of only -1.06%, compared to -1.45% for PiToMe and -4.62% for DSM. To our knowledge, AdaMerge is the first to combine salience-weighted similarity and adaptive per-layer reduction into a single training-free token merging framework, advancing the accuracy-FLOPs Pareto frontier of ViT acceleration.

arXiv:2605.27467v1 Announce Type: new Abstract: Traditional Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) units operate on discrete time steps, often failing to capture the fluid temporal dynamics of real-world physical processes. Liquid Neural Networks (LNNs), specifically Closed-form Continuous-time (CfC) networks, address this by modeling the hidden state evolution as a continuous differential equation. In this paper, we conduct a comprehensive benchmarking study across four distinct sequential modalities: neuromorphic event-based data (N-MNIST), stroke-based drawing (QuickDraw), visual handwriting (IAM), and physiological time-series (PhysioNet Sepsis-3). Furthermore, we perform a rigorous stress test using temporal dropout to evaluate model robustness against missing data. Our findings reveal that LNNs consistently provide superior parameter efficiency and significantly higher robustness in natively temporal domains and clinical environments where data sparsity is prevalent. This extended preprint provides additional background on related datasets and the LNN theoretical lineage, supplemented with a detailed appendix documenting our full implementation and experimental settings.

arXiv:2605.27487v1 Announce Type: new Abstract: Handwritten text generation (HTG) conditioned on writer style has been widely studied for Latin scripts, but remains underexplored for low-resource and non-Latin writing systems, leaving open how well existing models generalise beyond the Latin domain. Cyrillic, particularly Ukrainian, lacks both large-scale writer-labeled datasets and empirical evidence of such generalisation. To address this gap, we construct a Ukrainian handwritten word dataset of 126,177 images from 308 writers using connected-component segmentation, quality filtering, and targeted oversampling of underrepresented Ukrainian characters. We retrain DiffusionPen, a MobileNetV2 triplet-loss style encoder with a CANINE-conditioned latent diffusion U-Net, on this dataset without architectural modification, testing direct transfer from Latin to Cyrillic. We evaluate cross-domain style transfer in three settings: cross-lingual transfer from IAM English samples, zero-shot transfer to an early 20th-century Ukrainian manuscript, and few-shot imitation of contemporary writers. The model produces legible, style-consistent word images, indicating that few-shot latent diffusion models generalize beyond the Latin-script domain. We release the dataset, trained models, and evaluation protocol as a reproducible benchmark for writer-aware Cyrillic HTG, providing a foundation for extending stylized HTG to other underrepresented writing systems.

arXiv:2605.27469v1 Announce Type: new Abstract: Continual Learning (CL) is a practical paradigm to utilize power of deep pre-trained neural networks, but which pre-trained model has a better ability to balance ``Plasticity-Stability", deserving to be chosen? The logit shift serves as a natural proxy because it represents the logit shift in CL scenarios. However, obtaining the logit shift requires huge computational cost, which hinders large-scale model selection. Existing theoretical analyses fail to offer an efficient alternative because of the assumption of uniform hidden layer widths, which ignores the structural heterogeneity (variable width and depth) of real-world architectures. This raises a critical question: what theoretically relationship can be identified between heterogeneous architecture and logit shift on prior tasks (that the model has been trained on)? To answer the question, we decouple logit shift into architecture dependency and data dependency to establish our framework, which reveals that the combination of two dependency, defined as Architecture-driven Shift (ADS), that can capture the logit shift tendency well computable with few data samples. Specifically, for a well-optimized model on prior tasks, higher ADS is associated with a larger logit shift after training on the current task, which derived based on three mechanistic components: (1) spectral norm scaling of weight matrix gradients with layer width, (2) the optimization path length of the new task, and (3) the asymptotic task conflict in wide networks. Extensive empirical results across more than 175 diverse architectures demonstrate a strong monotonic correlation (the weakest Spearman's $r_s=0.731$) between ADS and logit shift. Practically, we demonstrate that ADS can serve as a lightweight proxy of the expected calibration error, which is a widely used metric for reliable CL model selection, on three datasets across six scenarios.

Discussion | Link

Michael Kan / PCMag: Filing: Amazon will acquire Apple's 20% stake in Globalstar as part of its acquisition of the satellite provider and create a subsidiary to conduct the merger  —  Amazon has revealed that it's taking over Apple's 20% stake in Globalstar as part of its $11.6 billion acquisition of the satellite communications provider.

Discussion | Link

Brian Danga / The Block: SoFi says SoFiUSD, its US dollar-pegged stablecoin on Ethereum and Solana, is now available for members to buy, sell, hold, and convert within the SoFi app  —  Quick Take  — SoFi said its U.S. dollar-pegged stablecoin is now available for members to buy, sell, hold, and convert within the SoFi app.

Given that the stock trading app operates in a highly regulated industry, the company’s move could prompt other finance firms to take a bold step and do the same.

The bill requires companies like OpenAI, Anthropic, and Google to have third parties confirm they’re following safety standards. Illinois governor JB Pritzker says he’ll sign it.

Jonathan Vanian / CNBC: Mark Zuckerberg tells shareholders that a Meta cloud computing business is “definitely on the table” if it overspends on data centers and has excess capacity  —  Meta CEO Mark Zuckerberg said his company could enter the cloud computing market if it overspends on data centers and ends up with excess capacity.

How to identify vulnerable borrowers and concentration risks before they become public problems

Explore OpenAI’s Frontier Governance Framework and how our AI safety, security, and risk practices align with emerging EU and California regulations.

<p><strong><a href="https://github.com/sqlite/sqlite/blob/master/AGENTS.md">sqlite AGENTS.md</a></strong></p> SQLite gained an AGENTS.md file <a href="https://github.com/sqlite/sqlite/commit/a1e5778889252d2609a59fd9b819d70392c5789e">five days ago</a> - but it's not intended for their own development, it's presumably aimed at people who are pointing agents at the SQLite codebase. It includes:</p> <blockquote> <p>SQLite does not accept pull requests without prior agreement and/or accompanying legal paperwork that places the pull request in the public domain. However, the human SQLite developers will review a concise and well-written pull request as a proof-of-concept prior to reimplementing the changes themselves.</p> <p>SQLite does not accept agentic code. However the project will accept agentic bug reports that include a reproducible test case. Patches or pull requests demonstrating a possible fix, for documentation purposes, are welcomed.</p> </blockquote> <p>The <a href="https://github.com/sqlite/sqlite/commit/db7fe319ed5a18dbc732ab8eacea557f41cd910f">most recent commit</a> to that file removed "(currently)" from "SQLite does not (currently) accept agentic code", with the commit message "Strengthen the statement about not accepting agentic code".</p> <p>Meanwhile the SQLite forum was being flooded with so many AI-generated bug reports - of varying quality - that they've now <a href="https://sqlite.org/forum/forumpost/2e7a8d6ba4b46d8315e80fd4a1e2feb40948dff5b7b11d5ba9cea5cb40aa252b">split those off</a> into a <a href="https://sqlite.org/bugs/forum">new SQLite Bug Forum</a>. D. Richard Hipp is resolving issues on there with a flurry of commits to the codebase. <p><small></small>Via <a href="https://discord.com/channels/823971286308356157/1097032579812687943/1507447792598253748">Alex Garcia on the Datasette Discord</a></small></p> <p>Tags: <a href="https://simonwillison.net/tags/sqlite">sqlite</a>, <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/d-richard-hipp">d-richard-hipp</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/coding-agents">coding-agents</a>, <a href="https://simonwillison.net/tags/ai-security-research">ai-security-research</a></p>

Jared Perlo / NBC News: Illinois Legislature passes SB 315, a bill requiring annual independent third-party safety audits of leading AI companies; the bill heads to the governor's desk  —  The measure, SB 315, goes beyond legislation from California and New York regulating America's largest AI companies.

Snowflake Inc.’s shares jumped in late trading today after it announced a $6 billion spending commitment on Amazon Web Services Inc.’s cloud infrastructure, including a deal to use the company’s custom artificial intelligence chips. The cloud data platform giant also reported stellar first-quarter earnings results, powering past Wall Street’s targets in a sign that the […] The post Snowflake’s stock surges after-hours on solid earnings beat and multibillion-dollar AWS cloud deal appeared first on SiliconANGLE.

Garrett Downs / CNBC: The US DOD announces a $9.7B five-year deal with Dell to provide Microsoft 365, advanced cloud subscriptions, and on-premises licensing to the US military  —  The Department of Defense on Wednesday announced a five-year, roughly $9.7 billion deal with Dell to provide a suite of software to the U.S. military.

Discussion | Link

ABC News: US prosecutors charge Google engineer Michele Spagnuolo with using inside information to make $1.2M on Polymarket bets about Google's Year in Search results  —  Michele Spagnuolo allegedly placed bets on what users were searching for.  —  In this Dec. 19, 2023, file photo …