Results for “model”

arXiv (research papers) Guijia Zhang SI

When Figures Disagree with Text: Evaluating Autonomous Visual Evidence Verification in Scientific Papers

Scientific figures carry evidence for the claims that cite them, so reviewing a paper means checking that a figure agrees with the text, and not only describing what the figure shows. Benchmarks of figure-text verification typically hand the model the figure, the claim, or the knowledge that an inconsistency exists, which leaves untested whether a model can find the figure worth checking in a complete paper, relate a visual element to the claim it conflicts with, and…

Read at arXiv (research papers)

arXiv (research papers) Marco Virgolin SI

automodel: A Skill for Agentic Model Discovery

The search for better models, whether expressed as analytical equations or neural architectures, has increasingly shifted from manual design toward data-driven automation. While a multitude of algorithms exist, ranging from exhaustive search to AlphaEvolve, recent advances in general-purpose coding agents such as Claude Code and OpenAI Codex raise the question: could providing such systems with a "model discovery recipe" be enough to discover or improve models? We present…

Read at arXiv (research papers)

arXiv (research papers) Kai Yi Ji SI

Can Weight-Space Programming Alter Learned Ethical Priorities?

We investigate whether a weight edit can alter how a model trades off honoring a commitment against benefiting more people. In a concept-supervised text model, we alter fitted choice weights so that the benefit required to override a formal pledge is larger than for an ordinary promise. Across six independently trained text readers, the edit selects the larger group in both orders on 100% of targeted ordinary conflicts and 83.3% of targeted formal conflicts. It preserves the…

Read at arXiv (research papers)

arXiv (research papers) Boxi Zhou SI

Reducing Inference Cost of Coding Agents with Timeout Cascades

AI coding agents have become one of the most popular applications of large language models. Yet, agentic AI comes at the cost of a tremendous amount of GPU compute resources, which limits its profitability and its viability at scale. In this paper, we investigate whether smaller, cheaper models can cut this cost without sacrificing accuracy. We evaluate 36 popular large and small open-weight models released since 2024 on a software engineering benchmark, and find that recent…

Read at arXiv (research papers)

arXiv (research papers) Zehao Gao SI

Fuse to Recognize the Unseen: Class-Relative Infrared-Visible Fusion for Open-Vocabulary Semantic Segmentation

Visible-infrared image fusion leverages complementary thermal cues to enhance visual perception, particularly under challenging illumination. However, such perceptual complementarity does not readily translate into semantic gains in open-vocabulary semantic segmentation (OVSS), where CLIP-based models are predominantly pretrained on RGB imagery. Directly introducing IR information may instead disrupt RGB-language alignment, turning promising thermal cues into semantic…

Read at arXiv (research papers)

arXiv (research papers) S. A. R. Horsley

Semiclassical Quantum Mechanics and the Optics of Time-Varying Media

Rapid, high contrast modulation of optical parameters such as the refractive index can radically reshape a wave's frequency spectrum as it propagates. Non-commuting frequency and time operators can be used to describe this reshaping, analogous to the position and momentum operators of quantum mechanics. Here we develop the connection between this theory and semi-classical quantum mechanics, showing that: (i) different microscopic models of time-varying materials are…

Read at arXiv (research papers)

MacRumors Tim Hardwick Gadgets

iPhone Air 2 Reportedly Sticks With Same Display

Apple's upcoming iPhone Air 2 will keep the same display specifications as its predecessor so the company can use existing OLED panel inventory, according to Korean publication ETNews. Citing industry sources, the report says new panel orders are expected to be minimal, with Samsung Display the sole supplier for the second-gen model. The report contrasts with earlier expectations that Apple would adopt Samsung's thinner, brighter display technology used in the iPhone Duo for…

Read at MacRumors

MacRumors Tim Hardwick Gadgets

iPhone 18 and iPhone Air 2 Display Production to Begin Soon Ahead of Expected March Launch

Apple's display suppliers are preparing to begin mass production of OLED panels for the upcoming iPhone Air 2 and iPhone 18, according to a report from Korea's ETNews. The devices will mark Apple's first spring releases next year as part of the company's new split-launch strategy for its flagship smartphones. Apple already debuted the iPhone 18 Pro models and iPhone Duo in September. The iPhone 18e, Apple's more affordable option, is also expected to launch in the spring…

Read at MacRumors

arXiv (research papers) Yavor Litchev Security

Retained Encrypted Search with Certified Split Joins

Retaining encrypted search across data edits and worker replacement requires separating computational validity from current release authority. We give exact Hamming search with at most one authorized decryption and decoding attempt per query preparation. Under authenticated prescribed execution, exclusive nonrollback owner state, an ideal public seed-expansion interface and a stated observation model, we reduce bounded adaptive privacy to ordinary ring learning with errors…

Read at arXiv (research papers)

arXiv (research papers) Tao Lian SI

FedE2E: One-Shot Federated Learning for End-to-End Autonomous Driving with Closed-Loop Evaluation

Federated learning enables collaborative model training without sharing datasets. However, existing federated learning methods for autonomous driving typically require multi-round communication and lack closed-loop evaluation, while one-shot methods mainly focus on perception tasks. To address these limitations, we propose \method, a one-shot federated learning method for end-to-end autonomous driving. It comprises multi-client calibration, anchor-based feature alignment…

Read at arXiv (research papers)

arXiv (research papers) Robin Geens

Hardware Acceleration of Vision-Language-Action Models with Weight Precision Rising over the Flow Steps

Vision-language-action (VLA) models such as GR00T control robots through an action head: a diffusion transformer that turns noise into a chunk of actions in a few flow-matching steps. Every step re-reads all of the head's weights from DRAM but applies each weight to only about fifty tokens. Inference is therefore memory-bound, and its latency and energy are largely determined by the total number of transferred weight bits. We show that later flow steps attenuate the…

Read at arXiv (research papers)

arXiv (research papers) Haoyang Liu Security

What Survives the Shift from Text to Pixels? Conditional Protection against Prompt Injection in Image-Rendered Documents

Multimodal language models increasingly answer questions over external documents, exposing them to indirect prompt injection: commands planted in untrusted content to hijack the response. Rendering untrusted content as page images lowers attack success on average, leaving open which injected operations survive rendering, at what utility cost, and whether failed image attacks reflect unread commands or commands read but not followed. We pair text and page-image versions of…

Read at arXiv (research papers)

arXiv (research papers) Samuel Fedida

A relativistic gravitisation of quantum theory?

We develop a model, that we call local semiclassical gravity, based on the distributional semiclassical Einstein field equations (EFEs) sourced by the local quantum state, i.e. the reduced density matrix of the local system updated by collapses (only) in its past light cone. We propose this as the basis of a relativistically causal generally covariant gravitisation of relativistic quantum theory in $1+3$ spacetime dimensions. We derive the weak-field limit which exhibits the…

Read at arXiv (research papers)

arXiv (research papers) Xiaoyang Xing SI

Learning to Coordinate: LLM-Guided Workflow Evolution for Token-Efficient Red Teaming

Automated red teaming is essential for identifying safety failures in large language models (LLMs), but its practical scalability depends on the resources required to discover those failures. Multi-agent systems can combine complementary capabilities, yet repeated planning, redundant exchanges, and ineffective information routing can make their coordination expensive. Moreover, a target model's response indicates the outcome of a test without identifying which collaboration…

Read at arXiv (research papers)

YourStory Startups

Sarvam’s move to host Chinese AI models triggers debate

Sarvam’s decision to host Chinese AI models on Indian infrastructure has ignited a debate over AI sovereignty, public funding and India's ambitions to develop homegrown frontier technology.

Read at YourStory

arXiv (research papers) Jun Li Security

Erased but Recoverable: On the Fragility of Concept Erasure Under Fine-Tuning

Existing concept erasure methods for text-to-image diffusion models primarily focus on suppressing target concepts while preserving non-target generation quality. However, in practical applications, open-source diffusion models are often further adapted through DreamBooth, LoRA, or Custom Diffusion. Such adaptations can reshape internal representations, thereby weakening or even reversing the erasure effects achieved by existing methods. We identify erasure persistence as a…

Read at arXiv (research papers)

arXiv (research papers) Johannes Hog SI

Towards Reliable Scaling Laws for Tabular Foundation Models

Scaling laws are widely used to predict foundation model performance and guide compute allocation. However, tabular foundation models (TFMs) differ from conventional foundation models in that they are pretrained on virtually unlimited synthetic data from configurable distributions, while ultimately being evaluated on real-world datasets. We investigate whether scaling laws for TabPFNv2-style models reliably extrapolate to larger compute budgets and generalize across…

Read at arXiv (research papers)

DW Geopolitics

Latin America's mega-prisons: Can El Salvador's model work?

El Salvador's mega-prison has become a model for governments across Latin America seeking to combat organized crime. But experts question whether the approach is effective, sustainable or compatible with human rights.

Read at DW

arXiv (research papers) Kevin Wardakhan SI

Conditional What-If Generation for Radio Access Network Energy under Rare and Unseen Contexts

Energy-saving mechanisms in RANs must balance energy reduction and network performance under varying traffic conditions. Conditional generative models support what-if analysis by generating plausible power-consumption trajectories for radio configurations and operating scenarios, helping operators evaluate energy-saving policies before deployment. However, representative measurements are costly and operationally constrained, so historical data cover only limited…

Read at arXiv (research papers)

arXiv (research papers) Bonan Ruan Security

Guarding the Kingdom of Skill-Based Agents in the AI Dynasty of Computation: Control-Flow Integrity for Agent Skills

Control-flow integrity (CFI) confines a program to a control-flow graph fixed ahead of time. Procedural agent skills move this control flow into prompt text, and a large language model (LLM) agent decides at runtime whether each step runs in its reasoning, a tool call, or a script. The procedure's control skeleton survives, but its enforceability depends on where each step runs. Many violations, such as overwriting a file before validation, use only operations the skill…

Read at arXiv (research papers)

Inc42 Palak Sharma Startups

Soket AI Raises $4 Mn From Bajaj Finserv To Expand Its AI Model Horizon

AI startup Soket AI has raised $4 Mn (around ₹37 Cr) from Bajaj Finserv’s applied research and innovation initiative Finserv…

Read at Inc42

SCMP China Dewey Sim Geopolitics

Carney hails ‘real progress’ in China-Canada ties but suggests US has no need to worry

Canadian Prime Minister Mark Carney said on Wednesday that ties with China were “growing on every front” after the two countries agreed to ease trade restrictions, but brushed off suggestions that was a threat to the United States. Carney also said he had raised the global governance of artificial intelligence when he met Chinese President Xi Jinping the previous day, setting out Canada’s “model for how this governance can be developed”. Relations between the two countries…

Read at SCMP China

TestingCatalog Alexey Shabanov SI

Google prepares to release Gemini 4 Neon model on APIs

Gemini 4 Neon appears in Google Cloud's SKU catalog at $3 and $15 per million tokens and in Antigravity CLI as a Pro orchestrator model.

Read at TestingCatalog

Infosecurity Magazine Security

Three AI Models Agree on Just One of 121 Potential Code Flaws

Echo tested Claude Mythos, GPT-5.6 Sol and GLM-5.3 on the same code and found little overlap

Read at Infosecurity Magazine

arXiv (research papers) Ernesto Biempica Tech policy

Human-LLM Cognitive Extension: A Framework for Studying Coupled Cognition

Recent research on the use of Large Language Models (LLMs) reports divergent cognitive outcomes, including improved assisted performance, limited transfer, and compromised unaided learning. Here, we argue that part of this seeming inconsistency arises because studies often assess different cognitive systems without making the unit of analysis explicit. Building on work in extended and distributed cognition, we propose the Human-LLM Cognitive Extension (HLCE) as a…

Read at arXiv (research papers)

arXiv (research papers) Long Xu Security

Evaluating LIM Trading Agents: A Critical Review of Backtesting Validity and Reporting Practices

Large language models (LLMs) are increasingly used in investment research, and many studies report promising historical returns. However, such results are only credible if researchers can demonstrate that the model used only information available at the time, account for how strategies and configurations were selected, and explain how reported returns were generated. We identify six common threats to validity in LLM investment evaluations and apply them to 18 published…

Read at arXiv (research papers)

The Register Carly Page SI

UK MPs want AI watchdog with power to pull the plug on dangerous models

Parliamentary committee says tech giants can't be left to mark their own homework when it comes to AI safety

Read at The Register

arXiv (research papers) Wen Zhang Security

PIVOT: Distribution-Preserving Multi-BitWatermarking with Minimax-Optimal Scoring

As LLM-generated content becomes increasingly widespread, reliable attribution calls for multi-bit watermarks that embed recoverable messages without distorting the output distribution. We introduce PIVOT, a multi-bit watermark with model-free verification based on mirrored rank-interval sampling. At each generation step, PIVOT uses a keyed pseudorandom scalar and a secret permutation of the vocabulary. The message bit determines whether next-token probability intervals are…

Read at arXiv (research papers)

arXiv (research papers) Marvin Qi

Stabilizer models of achiral abelian topological orders

We construct continuous-variable stabilizer codes realizing every achiral 2D abelian topological order, both bosonic and fermionic. Starting from an arbitrary signature-zero integer $K$-matrix, we show that the corresponding topological order can be obtained by condensing a lattice of dyons with trivial pairwise mutual braiding in an $\mathbb{R}^N$ gauge theory. We implement this condensation microscopically by gauging a non-on-site 1-form symmetry of the $\mathbb{R}^N$…

Read at arXiv (research papers)

Kyiv Post Kyiv Post

Ukrainian Battlefield Data Will Train Denmark’s AI

Danish armed forces will receive huge quantities of battlefield data and images from Ukraine with which it will train its own warfighting AI models. The Danish defense ministry announced that the two countries had signed an agreement on Tuesday.

Read at Kyiv Post

Older stories