Carlos's Debrief

⚡ Quick Summary

Agent Runtime Defense: ClawGuard treats indirect prompt injection as a systems problem for tool-augmented agents, proposing runtime protections around untrusted tool output. It stands out because production agents increasingly rely on web, local, and MCP-like inputs that can smuggle hostile instructions into the context window.
Cross-Trace Safety Auditing: One paper argues that rare failures like misuse campaigns, sabotage, reward hacking, and prompt injection are often only visible when auditors analyze many trajectories together. That matters because single-trace judges can miss the exact kinds of low-frequency failures that make deployed agents risky.
Reasoning Beyond Math: General365 expands reasoning evaluation beyond narrow specialist tasks into broader constraint-heavy, general-purpose problems. It matters because current leaderboard gains can overstate how well models generalize outside familiar benchmark regimes.
Simulators as Reasoning Data Engines: The Physics Olympiad RL paper uses physics simulators rather than internet QA pairs to train and test science reasoning. This is an important direction for domains where high-quality reasoning data is scarce but mechanistic environments are available.
GUI Agents Are Getting Infrastructure: ClawGUI focuses less on a single model and more on the missing full-stack training/evaluation pipeline for GUI agents. That matters because reliable automation over visual interfaces needs stable environments, evaluation discipline, and deployment tooling, not just bigger models.

📰 News Headlines

Supply-Chain Alarm: Coverage of a buyer backdooring dozens of WordPress plugins underscores how quickly trusted distribution channels can be weaponized once ownership changes.
Consumer AI Privacy Scrutiny: A reverse-engineering post claims ChatGPT's client waits for Cloudflare-side inspection of app state before enabling input, feeding a broader debate over how much telemetry AI frontends expose.
Frontier Cyber Eval Pressure: The UK AISI writeup says Claude Mythos Preview improved on CTF tasks and multi-step attack simulations, reinforcing the need for operationally realistic model capability testing.
Quantum + PQC Remain Hot: The web queue mixed post-quantum cryptography for IoT, quantum code-generation benchmarking, and even Rust-style safety ideas for quantum tooling, showing the stack is maturing beyond theory alone.
Agent Evaluation Is Broadening: New web papers emphasized unified agent benchmarks, persona-stable simulations, data lineage tracing, and RL credit assignment — all signs the field is shifting from flashy demos to harder reliability questions.

📄 ArXiv Papers

🧠 LLMs 9

A Mechanistic Analysis of Looped Reasoning Language Models [1]

Hugh Blayney, Álvaro Arroyo, Johan Obando-Ceron, Pablo Samuel Castro, et al.

Reasoning has become a central capability in large language models. Recent research has shown that reasoning performance can be improved by looping an LLM's layers in the latent dimension, resulting in looped reasoning language models. Why it matters: understanding where reasoning improves or fails remains central to…

cs.LGcs.AI

C-ReD: A Comprehensive Chinese Benchmark for AI-Generated Text Detection Derived from Real-World Prompts [2]

Chenxi Qing, Junxi Wu, Zheng Liu, Yixiang Qiu, et al.

Recently, large language models (LLMs) are capable of generating highly fluent textual content. While they offer significant convenience to humans, they also introduce various risks, like phishing and academic dishonesty. Why it matters: broader benchmarks shape what labs optimize for, and they expose where headline…

cs.CLcs.AI

General365: Benchmarking General Reasoning in Large Language Models Across Diverse and Challenging Tasks [3]

Junlin Liu, Shengnan An, Shuang Zhou, Dan Ma, et al.

Contemporary large language models (LLMs) have demonstrated remarkable reasoning capabilities, particularly in specialized domains like mathematics and physics. However, their ability to generalize these reasoning skills to more general and broader contexts--often termed general reasoning--remains under-explored. Why…

cs.CLcs.AI

OmniShow: Unifying Multimodal Conditions for Human-Object Interaction Video Generation [4]

Donghao Zhou, Guisheng Liu, Hao Yang, Jiatong Li, et al.

In this work, we study Human-Object Interaction Video Generation (HOIVG), which aims to synthesize high-quality human-object interaction videos conditioned on text, reference images, audio, and pose. This task holds significant practical value for automating content creation in real-world applications, such as…

cs.CV

Pair2Scene: Learning Local Object Relations for Procedural Scene Generation [5]

Xingjian Ran, Shujie Zhang, Weipeng Zhong, Li Luo, et al.

Generating high-fidelity 3D indoor scenes remains a significant challenge due to data scarcity and the complexity of modeling intricate spatial relations. Current methods often struggle to scale beyond training distribution to dense scenes or rely on LLMs/VLMs that lack the ability for precise spatial reasoning. Why…

cs.CV

Psychological Concept Neurons: Can Neural Control Bias Probing and Shift Generation in LLMs? [6]

Yuto Harada, Hiro Taiyo Hamada

Using psychological constructs such as the Big Five, large language models (LLMs) can imitate specific personality profiles and predict a user's personality. While LLMs can exhibit behaviors consistent with these constructs, it remains unclear where and how they are represented inside the model and how they relate to…

cs.CL

Saar-Voice: A Multi-Speaker Saarbrücken Dialect Speech Corpus [7]

Lena S. Oberkircher, Jesujoba O. Alabi, Dietrich Klakow, Jürgen Trouvain

Natural language processing (NLP) and speech technologies have made significant progress in recent years; however, they remain largely focused on standardized language varieties. Dialects, despite their cultural significance and widespread use, are underrepresented in linguistic resources and computational models…

cs.CL

Solving Physics Olympiad via Reinforcement Learning on Physics Simulators [8]

Mihir Prabhudesai, Aryan Satpathy, Yangmin Li, Zheyang Qin, et al.

We have witnessed remarkable advances in LLM reasoning capabilities with the advent of DeepSeek-R1. However, much of this progress has been fueled by the abundance of internet question-answer (QA) pairs, a major bottleneck going forward, since such data is limited in scale and concentrated mainly in domains like…

cs.LGcs.AIcs.CVcs.RO

Who Handles Orientation? Investigating Invariance in Feature Matching [9]

David Nordström, Johan Edstedt, Fredrik Kahl, Georg Bökman

Finding matching keypoints between images is a core problem in 3D computer vision. However, modern matchers struggle with large in-plane rotations. Why it matters: understanding where reasoning improves or fails remains central to building more capable and dependable language models.

cs.CV

🎨 Machine Learning 7

Budget-Aware Uncertainty for Radiotherapy Segmentation QA Using nnU-Net [10]

Ricardo Coimbra Brioso, Lorenzo Mondo, Damiano Dei, Nicola Lambri, et al.

Accurate delineation of the Clinical Target Volume (CTV) is essential for radiotherapy planning, yet remains time-consuming and difficult to assess, especially for complex treatments such as Total Marrow and Lymph Node Irradiation (TMLI). While deep learning-based auto-segmentation can reduce workload, safe clinical…

cs.CVcs.AI

Inferring Change Points in Regression via Sample Weighting [11]

Gabriel Arpino, Ramji Venkataramanan

We study the problem of identifying change points in high-dimensional generalized linear models, and propose an approach based on sample-weighted empirical risk minimization. Our method, Weighted ERM, encodes priors on the change points via weights assigned to each sample, to obtain weighted versions of standard…

stat.MEmath.STstat.ML

KL Divergence Between Gaussians: A Step-by-Step Derivation for the Variational Autoencoder Objective [12]

Andrés Muñoz, Rodrigo Ramele

Kullback-Leibler (KL) divergence is a fundamental concept in information theory that quantifies the discrepancy between two probability distributions. In the context of Variational Autoencoders (VAEs), it serves as a central regularization term, imposing structure on the latent space and thereby enabling the model to…

cs.LG

LangFlow: Continuous Diffusion Rivals Discrete in Language Modeling [13]

Yuxin Chen, Chumeng Liang, Hangke Sui, Ruihan Guo, et al.

Continuous diffusion models have achieved strong performance across domains such as images. However, in language modeling, prior continuous diffusion language models (DLMs) lag behind discrete counterparts. Why it matters: this contributes concrete data, benchmarks, or methods that can tighten the feedback loop…

cs.CLcs.LG

MosaicMRI: A Diverse Dataset and Benchmark for Raw Musculoskeletal MRI [14]

Paula Arguello, Berk Tinaz, Mohammad Shahab Sepehri, Maryam Soltanolkotabi, et al.

Deep learning underpins a wide range of applications in MRI, including reconstruction, artifact removal, and segmentation. However, progress has been driven largely by public datasets focused on brain and knee imaging, shaping how models are trained and evaluated. Why it matters: this contributes concrete data…

cs.CVcs.LGeess.SPphysics.med-ph

Nested Atoms Model with Application to Clustering Big Population-Scale Single-Cell Data [15]

Arhit Chakrabarti, Yang Ni, Yuchao Jiang, Bani K. Mallick

We consider the problem of clustering nested or hierarchical data, where observations are grouped and there are both group-level and observation-level variables. In our motivating OneK1K dataset, observations consist of single-cell RNA-sequencing (scRNA-seq) data from 982 individuals (groups), totaling 1.27 million…

stat.MEstat.APstat.ML

Physics-Informed State Space Models for Reliable Solar Irradiance Forecasting in Off-Grid Systems [16]

Mohammed Ezzaldin Babiker Abdullah

The stable operation of autonomous off-grid photovoltaic systems dictates reliance on solar forecasting algorithms that respect atmospheric thermodynamics. Contemporary deep learning models consistently exhibit critical anomalies, primarily severe temporal phase lags during cloud transients and physically impossible…

cs.LGcs.AIeess.SY

🔐 Security & Crypto 14

A Synthetic Conversational Smishing Dataset for Social Engineering Detection [17]

Carl Lochstampfor, Ayan Roy

Smishing (SMS phishing) has become a serious cybersecurity threat, especially for elderly and cyber-unaware individuals, causing financial loss and undermining user trust. Although prior work has focused on detecting smishing at the level of individual messages, real-world attackers often rely on multi-stage social…

cs.CR

AmBox: Device-to-Blockchain Ambient Sensing for Food Traceability [18]

João Miguel Guerreiro Fernandes, Samih Eisa, Miguel L. Pardal

From production to consumption, ensuring food quality and traceability depends on reliable monitoring of environmental conditions across the supply chain. Ambient sensing devices can collect relevant data such as temperature and humidity, but ensuring its integrity among stakeholders remains a challenge. Why it…

cs.CRcs.ET

ClawGuard: A Runtime Security Framework for Tool-Augmented LLM Agents Against Indirect Prompt Injection [19]

Wei Zhao, Zhe Li, Peixin Zhang, Jun Sun

Tool-augmented Large Language Model (LLM) agents have demonstrated impressive capabilities in automating complex, multi-step real-world tasks, yet remain vulnerable to indirect prompt injection. Adversaries exploit this weakness by embedding malicious instructions within tool-returned content, which agents directly…

cs.CRcs.AI

Detecting Safety Violations Across Many Agent Traces [20]

Adam Stein, Davis Brown, Hamed Hassani, Mayur Naik, et al.

To identify safety violations, auditors often search over large sets of agent traces. This search is difficult because failures are often rare, complex, and sometimes even adversarially hidden and only detectable when multiple traces are analyzed together. Why it matters: indirect prompt injection is one of the most…

cs.AIcs.CL

GPU Acceleration of Sparse Fully Homomorphic Encrypted DNNs [21]

Lara D'Agata, Carlos Agulló-Domingo, Óscar Vera-López, Kaustubh Shivdikar, et al.

Fully homomorphic encryption (FHE) has recently attracted significant attention as both a cryptographic primitive and a systems challenge. Given the latest advances in accelerated computing, FHE presents a promising opportunity for progress, with applications ranging from machine learning to information security. Why…

cs.CRcs.DCcs.DScs.LG

Hardware-Efficient Compound IC Protection with Lightweight Cryptography [22]

Levent Aksoy, Muhammad Sohaib Munir, Sedat Akleylek

Over the years, many techniques have been introduced to protect integrated circuits (ICs) from hardware security threats that emerged in the globalized IC manufacturing supply chain, such as overproduction and piracy. However, most of these techniques have been rendered inefficient since they do not rely on provably…

cs.CR

LMMs Meet Object-Centric Vision: Understanding, Segmentation, Editing and Generation [23]

Yuqian Yuan, Wenqiao Zhang, Juekai Lin, Yu Zhong, et al.

Large Multimodal Models (LMMs) have achieved remarkable progress in general-purpose vision--language understanding, yet they remain limited in tasks requiring precise object-level grounding, fine-grained spatial reasoning, and controllable visual manipulation. In particular, existing systems often struggle to identify…

cs.CV

Multi-ORFT: Stable Online Reinforcement Fine-Tuning for Multi-Agent Diffusion Planning in Cooperative Driving [24]

Haojie Bai, Aimin Li, Ruoyu Yao, Xiongwei Zhao, et al.

Closed-loop cooperative driving requires planners that generate realistic multimodal multi-agent trajectories while improving safety and traffic efficiency. Existing diffusion planners can model multimodal behaviors from demonstrations, but they often exhibit weak scene consistency and remain poorly aligned with…

cs.ROcs.AI

On the Robustness of Watermarking for Autoregressive Image Generation [25]

Andreas Müller, Denis Lukovnikov, Shingo Kodama, Minh Pham, et al.

The proliferation of autoregressive (AR) image generators demands reliable detection and attribution of their outputs to mitigate misinformation, and to filter synthetic images from training data to prevent model collapse. To address this need, watermarking techniques, specifically designed for AR models, embed a…

cs.CVcs.AIcs.CR

Public Key Encryption from High-Corruption Constraint Satisfaction Problems [26]

Isaac M Hair, Amit Sahai

We give a public key encryption scheme with plausible quasi-exponential security based on the conjectured intractability of two constraint satisfaction problems (CSPs), both of which are instantiated with a corruption rate of $1 - o(1)$. First, we conjecture the hardness of a new large alphabet random predicate CSP…

cs.CR

RLSpoofer: A Lightweight Evaluator for LLM Watermark Spoofing Resilience [27]

Hanbo Huang, Xuan Gong, Yiran Zhang, Hao Zheng, et al.

Large language model (LLM) watermarking has emerged as a promising approach for detecting and attributing AI-generated text, yet its robustness to black-box spoofing remains insufficiently evaluated. Existing evaluation methods often demand extensive datasets and white-box access to algorithmic internals, limiting…

cs.CR

Security Implications of 5G Communication in Industrial Systems [28]

Stefan Lenz, Sotiris Michaelides, Moritz Rickert, Jonas Holtwick, et al.

Traditionally, industrial control systems (ICS) were designed without security in mind, prioritizing availability and real-time communication. As these systems increasingly become targets of powerful adversaries, security can no longer be neglected. Why it matters: security papers are increasingly focused on making…

cs.CRcs.NIeess.SY

Semiprojective Banach lattices [29]

Tomasz Kania, Mariusz Niwiński

We introduce a norm-controlled notion of semiprojectivity for Banach lattices, requiring liftability of contractive lattice homomorphisms through inductive limits of closed ideals with arbitrarily small loss of norm control. Our main result establishes that, for a compact metric space $X$, the Banach lattice $C(X)$ is…

math.FAmath.CTmath.OA

Towards Automated Pentesting with Large Language Models [30]

Ricardo Bessa, Rui Claro, João Trindade, João Lourenço

Large Language Models (LLMs) are redefining offensive cybersecurity by allowing the generation of harmful machine code with minimal human intervention. While attackers take advantage of dark LLMs such as XXXGPT and WolfGPT to produce malicious code, ethical hackers can follow similar approaches to automate traditional…

cs.CR

🕵️ Zero Knowledge 1

Optimal Codes for Deterministic Identification over Gaussian Channels: Closing the Capacity Gap [31]

Pau Colomer, Christian Deppe, Holger Boche, Andreas Winter

Deterministic identification (DI) has emerged as a promising paradigm for large-scale and goal-oriented communication systems. Despite significant progress, a fundamental open problem has remained unresolved: a persistent gap between the best known lower and upper bounds on the DI capacity, as well as on the…

cs.IT

⚛️ Quantum 3

Extraction of Effective Electromagnetic Material Properties for Rydberg Electrometer Vapor Cells from 10-300 MHz [32]

D. Richardson, J. Dee, J. Yaeger, M. Viray, et al.

Quantum sensors often consist of packaging, such as dielectric-based vapor cells and metallic electrodes, that reduces and spatially alters the locally observed electromagnetic fields. These effects have been well studied in the optical regime, and even in the RF regime over a few GHz. Why it matters: quantum-adjacent…

physics.atom-ph

Generalized Roth--Lempel Codes: NMDS Characterization, Hermitian Self-Orthogonality, and Quantum Constructions [33]

Qi Liu, Xuefei Wu, Haiyan Zhou

In their seminal 1989 work (IEEE Trans. Inf. Why it matters: quantum-adjacent tooling is still immature, so benchmarks and systems work can meaningfully shape the field’s practical direction.

cs.IT

Many-Body Super- and Subradiance in Ordered Atomic Arrays [34]

Alec Douglas, Lin Su, Michal Szurek, Robin Groth, et al.

When quantum emitters couple indistinguishably to light, they can synchronize into a collective light matter system with radiative properties profoundly different from those of independent particles. To date, the resulting collective effects have largely been confined to point like or homogeneous ensembles. Why it…

quant-phcond-mat.quant-gasphysics.atom-ph

🛡️ AI Safety 2

Autonomous Diffractometry Enabled by Visual Reinforcement Learning [35]

J. Oppliger, M. Stifter, A. Rüegg, I. Biało, et al.

Automation underpins progress across scientific and industrial disciplines. Yet, automating tasks requiring interpretation of abstract visual information remain challenging. Why it matters: safety papers that improve auditing or calibrated decisions are increasingly important as more model behavior gets delegated to…

cs.LGcond-mat.mtrl-scics.CV

CLSGen: A Dual-Head Fine-Tuning Framework for Joint Probabilistic Classification and Verbalized Explanation [36]

WonJin Yoon, Kangyu Zhu, Ian Bulovic, Autumn Sehy, et al.

With the recent progress of Large Language Models (LLMs), there is a growing interest in applying these models to solve complex and challenging problems. Modern LLMs, capable of processing long contexts and generating verbalized explanations, offer significant potential in addressing real-world applications. Why it…

cs.CL

🤖 Agents 3

ClawGUI: A Unified Framework for Training, Evaluating, and Deploying GUI Agents [37]

Fei Tang, Zhiqiong Lu, Boxuan Zhang, Weiming Lu, et al.

GUI agents drive applications through their visual interfaces instead of programmatic APIs, interacting with arbitrary software via taps, swipes, and keystrokes, reaching a long tail of applications that CLI-based agents cannot. Yet progress in this area is bottlenecked less by modeling capacity than by the absence of…

cs.LGcs.AIcs.CLcs.CV

Disentangled Point Diffusion for Precise Object Placement [38]

Lyuxing He, Eric Cai, Shobhit Aggarwal, Jianjun Wang, et al.

Recent advances in robotic manipulation have highlighted the effectiveness of learning from demonstration. However, while end-to-end policies excel in expressivity and flexibility, they struggle both in generalizing to novel object geometries and in attaining a high degree of precision. Why it matters: agent research…

cs.RO

GenTac: Generative Modeling and Forecasting of Soccer Tactics [39]

Jiayuan Rao, Tianlin Gui, Haoning Wu, Yanfeng Wang, et al.

Modeling open-play soccer tactics is a formidable challenge due to the stochastic, multi-agent nature of the game. Existing computational approaches typically produce single, deterministic trajectory forecasts or focus on highly structured set-pieces, fundamentally failing to capture the inherent variance and…

cs.AIcs.MA

🌐 Web Findings

🦞 Lobste.rs 7

ChatGPT Won't Let You Type Until Cloudflare Reads Your React State. I Decrypted the Program That Does It [40]

Lobste.rs discussion • 2026-03-30

Edit April 2, 2026: I've been getting inbound interest from researchers wanting to run their own queries. The MCP integration I use for my own research lets you analyze live mobile telemetry continuously collected from real devices in the wild, directly from…

🦞 Lobste.rsprivacy2026-03-30

LARQL - Query neural network weights like a graph database [41]

Lobste.rs discussion • 2026-04-14

Contribute to chrishayuk/larql development by creating an account on GitHub.

🦞 Lobste.rsai2026-04-14

No one can force me to have a secure website [42]

Lobste.rs discussion • 2026-04-13

Additional context: https://www.youtube.com/watch?v=M1si1y5lvkk No abstract.

🦞 Lobste.rssecurity2026-04-13

Our evaluation of Claude Mythos Preview’s cyber capabilities [43]

Lobste.rs discussion • 2026-04-14

We conducted cyber evaluations of Anthropic’s Claude Mythos Preview and found continued improvement in capture-the-flag (CTF) challenges and significant improvement on multi-step cyber-attack simulations.

🦞 Lobste.rssecurity2026-04-14

Someone Bought 30 WordPress Plugins and Planted a Backdoor in All of Them [44]

Lobste.rs discussion • 2026-04-13

Last week, I wrote about catching a supply chain attack on a WordPress plugin called Widget Logic. A trusted name, acquired by a new owner, turned into

🦞 Lobste.rssecurity2026-04-13

TESSERA — A pixel-wise earth observation foundation model [45]

Lobste.rs discussion • 2026-04-14

TESSERA compresses a year of satellite imagery into dense per-pixel embeddings at 10m resolution. Open data, open weights, open embeddings.

🦞 Lobste.rsai2026-04-14

The Origins of GPU Computing [46]

Lobste.rs discussion • 2026-04-13

Historical article tracing how graphics hardware evolved into the modern GPU-computing stack that now underpins large-scale AI workloads.

🦞 Lobste.rsai2026-04-13

🎓 Google Scholar 8

AI and Cybersecurity: Opportunities, challenges, and governance [47]

Google Scholar result

… as adversarial AI and ethical dilemmas. This paper studies AI’s transformative role in cybersecurity, … We examine AI’s applications in threat detection, incident response, and …

🎓 Google ScholarAI security cybersecurity

Artificial intelligence and machine learning in cybersecurity: a deep dive into state-of-the-art techniques and future paradigms [48]

Google Scholar result

… in adversarial AI, automated threat intelligence, and AI-driven security orchestration, this … AI’s role in cybersecurity. Figure 1 shows the key areas where Artificial intelligence (AI) and …

🎓 Google ScholarAI security cybersecurity

Artificial intelligence research in tourism and hospitality journals: Trends, emerging themes, and the rise of generative AI [49]

Google Scholar result

… themes of artificial intelligence in the field of tourism and hospitality research. On 5 March … This study on artificial intelligence in tourism and hospitality research highlighted significant …

🎓 Google Scholarartificial intelligence research 2025

Catalyst breakthroughs in methane dry reforming: Employing machine learning for future advancements [50]

Google Scholar result

Rising levels of atmospheric carbon dioxide (CO 2 ) and methane (CH 4 ) have sparked the interest of researchers in resolving this issue. Various technologies have been utilized such …

🎓 Google Scholarmachine learning breakthroughs

Early identification of breakthrough technologies: Insights from science-driven innovations [51]

Google Scholar result

… Building on this premise, this paper presents a framework for identifying breakthrough … potential to trigger technological breakthroughs. Next, a machine learning-based link prediction …

🎓 Google Scholarmachine learning breakthroughs

Large language models (LLM) in computational social science: prospects, current state, and challenges [52]

Google Scholar result

… The advent of large language models (LLMs) has marked a new … LLM usage. We further present the challenges associated with data bias, privacy, and the integration of these models …

🎓 Google Scholarlarge language models LLM

Quantum machine learning: A comprehensive review of integrating AI with quantum computing for computational advancements [53]

Google Scholar result

… in quantum-enhanced classical ML to native quantum algorithms and hybrid quantum-… It varies from applications in optimization, drug discovery, and quantum-secured communications, …

🎓 Google Scholarquantum computing algorithms

Securing the future: exploring post-quantum cryptography for authentication and user privacy in IoT devices [54]

Google Scholar result

Cluster Computing - With the emergence of quantum computers, traditional cryptographic methods are vulnerable to attacks, emphasizing the need for post-quantum cryptography to secure devices and...

🎓 Google Scholarcryptography post-quantum

🗞️ Hacker News 1

Show HN: B01T – What if quantum computing had Rust-style safety guarantees? [55]

Hacker News discussion

Hacker News launch post for B01T – What if quantum computing had Rust-style safety guarantees?, a project exploring safer abstractions and developer ergonomics for quantum computing.

🗞️ Hacker Newsquantum computing

🤗 HuggingFace 30

ADD for Multi-Bit Image Watermarking [56]

An Luo, Jie Ding

As generative models enable rapid creation of high-fidelity images, societal concerns about misinformation and authenticity have intensified. A promising remedy is multi-bit image watermarking, which embeds a multi-bit message into an image so that a verifier…

🤗 HuggingFacedaily curated papers2026-04-13

Advancing Polish Language Modeling through Tokenizer Optimization in the Bielik v3 7B and 11B Series [57]

Krzysztof Ociepa, Łukasz Flis, Remigiusz Kinas, Krzysztof Wróbel, et al.

The development of the Bielik v3 PL series, encompassing both the 7B and 11B parameter variants, represents a significant milestone in the field of language-specific large language model (LLM) optimization. While general-purpose models often demonstrate…

🤗 HuggingFacedaily curated papers2026-04-12

Attention Sink in Transformers: A Survey on Utilization, Interpretation, and Mitigation [58]

Zunhai Su, Hengyuan Zhang, Wei Wu, Yifan Zhang, et al.

As the foundational architecture of modern machine learning, Transformers have driven remarkable progress across diverse AI domains. Despite their transformative impact, a persistent challenge across various Transformers is Attention Sink (AS), in which a…

🤗 HuggingFacedaily curated papers2026-04-11

Audio Flamingo Next: Next-Generation Open Audio-Language Models for Speech, Sound, and Music [59]

Sreyan Ghosh, Arushi Goel, Kaousheik Jayakumar, Lasha Koroshinadze, et al.

We present Audio Flamingo Next (AF-Next), the next-generation and most capable large audio-language model in the Audio Flamingo series, designed to advance understanding and reasoning over speech, environmental sounds and music. Compared to Audio Flamingo 3…

🤗 HuggingFacedaily curated papers2026-04-13

BMdataset: A Musicologically Curated LilyPond Dataset [60]

Matteo Spanio, Ilay Guler, Antonio Rodà

Symbolic music research has relied almost exclusively on MIDI-based datasets; text-based engraving formats such as LilyPond remain unexplored for music understanding. We present BMdataset, a musicologically curated dataset of 393 LilyPond scores (2,646…

🤗 HuggingFacedaily curated papers2026-04-12

CocoaBench: Evaluating Unified Digital Agents in the Wild [61]

CocoaBench Team, Shibo Hao, Zhining Zhang, Zhiqi Liang, et al.

LLM agents now perform strongly in software engineering, deep research, GUI automation, and various other applications, while recent agent scaffolds and models are increasingly integrating these capabilities into unified systems. Yet, most evaluations still…

🤗 HuggingFacedaily curated papers2026-04-13

CodeTracer: Towards Traceable Agent States [62]

Han Li, Yifan Yao, Letian Zhu, Rili Feng, et al.

Code agents are advancing rapidly, but debugging them is becoming increasingly difficult. As frameworks orchestrate parallel tool calls and multi-stage workflows over complex tasks, making the agent's state transitions and error propagation hard to observe.

🤗 HuggingFacedaily curated papers2026-04-13

Continuous Adversarial Flow Models [63]

Shanchuan Lin, Ceyuan Yang, Zhijie Lin, Hao Chen, et al.

We propose continuous adversarial flow models, a type of continuous-time flow model trained with an adversarial objective. Unlike flow matching, which uses a fixed mean-squared-error criterion, our approach introduces a learned discriminator to guide training.

🤗 HuggingFacedaily curated papers2026-04-13

DiningBench: A Hierarchical Multi-view Benchmark for Perception and Reasoning in the Dietary Domain [64]

Song Jin, Juntian Zhang, Xun Zhang, Zeying Tian, et al.

Recent advancements in Vision-Language Models (VLMs) have revolutionized general visual understanding. However, their application in the food domain remains constrained by benchmarks that rely on coarse-grained categories, single-view imagery, and inaccurate…

🤗 HuggingFacedaily curated papers2026-04-12

Eliciting Medical Reasoning with Knowledge-enhanced Data Synthesis: A Semi-Supervised Reinforcement Learning Approach [65]

Haolin Li, Shuyang Jiang, Ruipeng Zhang, Jiangchao Yao, et al.

While large language models hold promise for complex medical applications, their development is hindered by the scarcity of high-quality reasoning data. To address this issue, existing approaches typically distill chain-of-thought reasoning traces from large…

🤗 HuggingFacedaily curated papers2026-04-13

From Reasoning to Agentic: Credit Assignment in Reinforcement Learning for Large Language Models [66]

Chenchen Zhang

Reinforcement learning (RL) for large language models (LLMs) increasingly relies on sparse, outcome-level rewards -- yet determining which actions within a long trajectory caused the outcome remains difficult. This credit assignment (CA) problem manifests in…

🤗 HuggingFacedaily curated papers2026-04-13

Introspective Diffusion Language Models [67]

Yifan Yu, Yuqing Jian, Junxiong Wang, Zhongzhu Zhou, et al.

Diffusion language models promise parallel generation, yet still lag behind autoregressive (AR) models in quality. We stem this gap to a failure of introspective consistency: AR models agree with their own generations, while DLMs often do not.

🤗 HuggingFacedaily curated papers2026-04-13

Low-rank Optimization Trajectories Modeling for LLM RLVR Acceleration [68]

Zhipeng Chen, Tao Qian, Wayne Xin Zhao, Ji-Rong Wen

Recently, scaling reinforcement learning with verifiable rewards (RLVR) for large language models (LLMs) has emerged as an effective training paradigm for significantly improving model capabilities, which requires guiding the model to perform extensive…

🤗 HuggingFacedaily curated papers2026-04-13

Not All Denoising Steps Are Equal: Model Scheduling for Faster Masked Diffusion Language Models [69]

Ivan Sedykh, Nikita Sorokin, Valentin Malykh

Recent advances in masked diffusion language models (MDLMs) narrow the quality gap to autoregressive LMs, but their sampling remains expensive because generation requires many full-sequence denoising passes with a large Transformer and, unlike autoregressive…

🤗 HuggingFacedaily curated papers2026-04-11

Playing Along: Learning a Double-Agent Defender for Belief Steering via Theory of Mind [70]

Hanqi Xiao, Vaidehi Patil, Zaid Khan, Hyunji Lee, et al.

As large language models (LLMs) become the engine behind conversational systems, their ability to reason about the intentions and states of their dialogue partners (i.e., form and use a theory-of-mind, or ToM) becomes increasingly critical for safe…

🤗 HuggingFacedaily curated papers2026-04-13

Polyglot Teachers: Evaluating Language Models for Multilingual Synthetic Data Generation [71]

Lester James V. Miranda, Ivan Vulić, Anna Korhonen

Synthesizing supervised finetuning (SFT) data from language models (LMs) to teach smaller models multilingual tasks has become increasingly common. However, teacher model selection is often ad hoc, typically defaulting to the largest available option, even…

🤗 HuggingFacedaily curated papers2026-04-13

Prompt Relay: Inference-Time Temporal Control for Multi-Event Video Generation [72]

Gordon Chen, Ziqi Huang, Ziwei Liu

Video diffusion models have achieved remarkable progress in generating high-quality videos. However, these models struggle to represent the temporal succession of multiple events in real-world videos and lack explicit mechanisms to control when semantic…

🤗 HuggingFacedaily curated papers2026-04-11

Pseudo-Unification: Entropy Probing Reveals Divergent Information Patterns in Unified Multimodal Models [73]

Songlin Yang, Xianghao Kong, Anyi Rao

Unified multimodal models (UMMs) were designed to combine the reasoning ability of large language models (LLMs) with the generation capability of vision models. In practice, however, this synergy remains elusive: UMMs fail to transfer LLM-like reasoning to…

🤗 HuggingFacedaily curated papers2026-04-13

QuanBench+: A Unified Multi-Framework Benchmark for LLM-Based Quantum Code Generation [74]

Ali Slim, Haydar Hamieh, Jawad Kotaich, Yehya Ghosn, et al.

Large Language Models (LLMs) are increasingly used for code generation, yet quantum code generation is still evaluated mostly within single frameworks, making it difficult to separate quantum reasoning from framework familiarity. We introduce QuanBench+, a…

🤗 HuggingFacedaily curated papers2026-03-25

ScheMatiQ: From Research Question to Structured Data through Interactive Schema Discovery [75]

Shahar Levy, Eliya Habba, Reshef Mintz, Barak Raveh, et al.

Many disciplines pose natural-language research questions over large document collections whose answers typically require structured evidence, traditionally obtained by manually designing an annotation schema and exhaustively labeling the corpus, a slow and…

🤗 HuggingFacedaily curated papers2026-04-10

SciPredict: Can LLMs Predict the Outcomes of Scientific Experiments in Natural Sciences? [76]

Udari Madhushani Sehwag, Elaine Lau, Haniyeh Ehsani Oskouie, Shayan Shabihi, et al.

Accelerating scientific discovery requires the identification of which experiments would yield the best outcomes before committing resources to costly physical validation. While existing benchmarks evaluate LLMs on scientific knowledge and reasoning, their…

🤗 HuggingFacedaily curated papers2026-04-12

SPASM: Stable Persona-driven Agent Simulation for Multi-turn Dialogue Generation [77]

Han Luo, Guy Laban

Large language models are increasingly deployed in multi-turn settings such as tutoring, support, and counseling, where reliability depends on preserving consistent roles, personas, and goals across long horizons. This requirement becomes critical when LLMs…

🤗 HuggingFacedaily curated papers2026-04-10

SPEED-Bench: A Unified and Diverse Benchmark for Speculative Decoding [78]

Talor Abramovich, Maor Ashkenazi, Carl, Putterman, et al.

Speculative Decoding (SD) has emerged as a critical technique for accelerating Large Language Model (LLM) inference. Unlike deterministic system optimizations, SD performance is inherently data-dependent, meaning that diverse and representative workloads are…

🤗 HuggingFacedaily curated papers2026-02-10

Strips as Tokens: Artist Mesh Generation with Native UV Segmentation [79]

Rui Xu, Dafei Qin, Kaichun Qiao, Qiujie Dong, et al.

Recent advancements in autoregressive transformers have demonstrated remarkable potential for generating artist-quality meshes. However, the token ordering strategies employed by existing methods typically fail to meet professional artist standards, where…

🤗 HuggingFacedaily curated papers2026-04-10

SWE-AGILE: A Software Agent Framework for Efficiently Managing Dynamic Reasoning Context [80]

Shuquan Lian, Juncheng Liu, Yazhe Chen, Yuhong Chen, et al.

Prior representative ReAct-style approaches in autonomous Software Engineering (SWE) typically lack the explicit System-2 reasoning required for deep analysis and handling complex edge cases. While recent reasoning models demonstrate the potential of extended…

🤗 HuggingFacedaily curated papers2026-04-13

TAIHRI: Task-Aware 3D Human Keypoints Localization for Close-Range Human-Robot Interaction [81]

Ao Li, Yonggen Ling, Yiyang Lin, Yuji Wang, et al.

Accurate 3D human keypoints localization is a critical technology enabling robots to achieve natural and safe physical interaction with users. Conventional 3D human keypoints estimation methods primarily focus on the whole-body reconstruction quality relative…

🤗 HuggingFacedaily curated papers2026-04-10

TorchUMM: A Unified Multimodal Model Codebase for Evaluation, Analysis, and Post-training [82]

Yinyi Luo, Wenwen Wang, Hayes Bai, Hongyu Zhu, et al.

Recent advances in unified multimodal models (UMMs) have led to a proliferation of architectures capable of understanding, generating, and editing across visual and textual modalities. However, developing a unified framework for UMMs remains challenging due…

🤗 HuggingFacedaily curated papers2026-04-12

Tracing the Roots: A Multi-Agent Framework for Uncovering Data Lineage in Post-Training LLMs [83]

Yu Li, Xiaoran Shang, Qizhi Pei, Yun Zhu, et al.

Post-training data plays a pivotal role in shaping the capabilities of Large Language Models (LLMs), yet datasets are often treated as isolated artifacts, overlooking the systemic connections that underlie their evolution. To disentangle these complex…

🤗 HuggingFacedaily curated papers2026-04-12

Uni-ViGU: Towards Unified Video Generation and Understanding via A Diffusion-Based Video Generator [84]

Luozheng Qin, Jia Gong, Qian Qiao, Tianjiao Li, et al.

Unified multimodal models integrating visual understanding and generation face a fundamental challenge: visual generation incurs substantially higher computational costs than understanding, particularly for video. This imbalance motivates us to invert the…

🤗 HuggingFacedaily curated papers2026-04-09

Zero-shot World Models Are Developmentally Efficient Learners [85]

Khai Loong Aw, Klemen Kotar, Wanhee Lee, Seungwoo Kim, et al.

Young children demonstrate early abilities to understand their physical world, estimating depth, motion, object coherence, interactions, and many other aspects of physical scene understanding. Children are both data-efficient and flexible cognitive systems…

🤗 HuggingFacedaily curated papers2026-04-11

⚡ Quick Summary

📰 News Headlines

📄 ArXiv Papers

🧠 LLMs 9

🎨 Machine Learning 7

🔐 Security & Crypto 14

🕵️ Zero Knowledge 1

⚛️ Quantum 3

🛡️ AI Safety 2

🤖 Agents 3

🌐 Web Findings

🦞 Lobste.rs 7

🎓 Google Scholar 8

🗞️ Hacker News 1

🤗 HuggingFace 30

🔗 All Sources