Learning to Stay Safe: Adaptive Regularization Against Safety Degradation during Fine-Tuning
arXiv:2602.17546v1 Announce Type: new Abstract: Instruction-following language models are trained to be helpful and safe, yet their safety behavior can deteriorate under benign fine-tuning and worsen under adversarial updates. Existing defenses often offer limited protection or force a trade-off between...
Differences in Typological Alignment in Language Models' Treatment of Differential Argument Marking
arXiv:2602.17653v1 Announce Type: new Abstract: Recent work has shown that language models (LMs) trained on synthetic corpora can exhibit typological preferences that resemble cross-linguistic regularities in human languages, particularly for syntactic phenomena such as word order. In this paper, we...
Intent Laundering: AI Safety Datasets Are Not What They Seem
arXiv:2602.16729v1 Announce Type: cross Abstract: We systematically evaluate the quality of widely used AI safety datasets from two perspectives: in isolation and in practice. In isolation, we examine how well these datasets reflect real-world attacks based on three key properties:...
Hybrid-Gym: Training Coding Agents to Generalize Across Tasks
arXiv:2602.16819v1 Announce Type: cross Abstract: When assessing the quality of coding agents, predominant benchmarks focus on solving single issues on GitHub, such as SWE-Bench. In contrast, in real use, these agents solve more various and complex tasks that involve other...
DeepVision-103K: A Visually Diverse, Broad-Coverage, and Verifiable Mathematical Dataset for Multimodal Reasoning
arXiv:2602.16742v1 Announce Type: new Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has been shown effective in enhancing the visual reflection and reasoning capabilities of Large Multimodal Models (LMMs). However, existing datasets are predominantly derived from either small-scale manual construction or...
Escaping the Cognitive Well: Efficient Competition Math with Off-the-Shelf Models
arXiv:2602.16793v1 Announce Type: new Abstract: In the past year, custom and unreleased math reasoning models reached gold medal performance on the International Mathematical Olympiad (IMO). Similar performance was then reported using large-scale inference on publicly available models but at prohibitive...
HiVAE: Hierarchical Latent Variables for Scalable Theory of Mind
arXiv:2602.16826v1 Announce Type: new Abstract: Theory of mind (ToM) enables AI systems to infer agents' hidden goals and mental states, but existing approaches focus mainly on small human understandable gridworld spaces. We introduce HiVAE, a hierarchical variational architecture that scales...
Training Large Reasoning Models Efficiently via Progressive Thought Encoding
arXiv:2602.16839v1 Announce Type: new Abstract: Large reasoning models (LRMs) excel on complex problems but face a critical barrier to efficiency: reinforcement learning (RL) training requires long rollouts for outcome-based rewards, where autoregressive decoding dominates time and memory usage. While sliding-window...
Position: Why a Dynamical Systems Perspective is Needed to Advance Time Series Modeling
arXiv:2602.16864v1 Announce Type: new Abstract: Time series (TS) modeling has come a long way from early statistical, mainly linear, approaches to the current trend in TS foundation models. With a lot of hype and industrial demand in this field, it...
A Unified Framework for Locality in Scalable MARL
arXiv:2602.16966v1 Announce Type: new Abstract: Scalable Multi-Agent Reinforcement Learning (MARL) is fundamentally challenged by the curse of dimensionality. A common solution is to exploit locality, which hinges on an Exponential Decay Property (EDP) of the value function. However, existing conditions...
Early-Warning Signals of Grokking via Loss-Landscape Geometry
arXiv:2602.16967v1 Announce Type: new Abstract: Grokking -- the abrupt transition from memorization to generalization after prolonged training -- has been linked to confinement on low-dimensional execution manifolds in modular arithmetic. Whether this mechanism extends beyond arithmetic remains open. We study...
Discovering Universal Activation Directions for PII Leakage in Language Models
arXiv:2602.16980v1 Announce Type: new Abstract: Modern language models exhibit rich internal structure, yet little is known about how privacy-sensitive behaviors, such as personally identifiable information (PII) leakage, are represented and modulated within their hidden states. We present UniLeak, a mechanistic-interpretability...
Action-Graph Policies: Learning Action Co-dependencies in Multi-Agent Reinforcement Learning
arXiv:2602.17009v1 Announce Type: new Abstract: Coordinating actions is the most fundamental form of cooperation in multi-agent reinforcement learning (MARL). Successful decentralized decision-making often depends not only on good individual actions, but on selecting compatible actions across agents to synchronize behavior,...
WS-GRPO: Weakly-Supervised Group-Relative Policy Optimization for Rollout-Efficient Reasoning
arXiv:2602.17025v1 Announce Type: new Abstract: Group Relative Policy Optimization (GRPO) is effective for training language models on complex reasoning. However, since the objective is defined relative to a group of sampled trajectories, extended deliberation can create more chances to realize...
Sign Lock-In: Randomly Initialized Weight Signs Persist and Bottleneck Sub-Bit Model Compression
arXiv:2602.17063v1 Announce Type: new Abstract: Sub-bit model compression seeks storage below one bit per weight; as magnitudes are aggressively compressed, the sign bit becomes a fixed-cost bottleneck. Across Transformers, CNNs, and MLPs, learned sign matrices resist low-rank approximation and are...
Spatio-temporal dual-stage hypergraph MARL for human-centric multimodal corridor traffic signal control
arXiv:2602.17068v1 Announce Type: new Abstract: Human-centric traffic signal control in corridor networks must increasingly account for multimodal travelers, particularly high-occupancy public transportation, rather than focusing solely on vehicle-centric performance. This paper proposes STDSH-MARL (Spatio-Temporal Dual-Stage Hypergraph based Multi-Agent Reinforcement Learning),...
India’s Sarvam launches Indus AI chat app as competition heats up
Sarvam's Indus chat app is currently available in beta.
Why creators are ditching ad revenue for chocolate bars and fintech acquisitions
The creator economy is evolving fast, and ad revenue alone isn’t cutting it anymore. YouTubers are launching product lines, acquiring startups, and building actual business empires. In fact, MrBeast’s company bought fintech startup Step, and his chocolate business is out-earning...
InScope nabs $14.5M to solve the pain of financial reporting
The startup, founded by accountants who worked at Flexport, Miro, Hopin and Thrive Global, automates the difficulties of prepping financial statements.
Great news for xAI: Grok is now pretty good at answering questions about Baldur’s Gate
A new report from Business Insider reveals that high-level engineers at xAI were pulled off other projects to make sure Grok could answer detailed questions about the video game Baldur's Gate.
‘Toy Story 5’ takes aim at creepy AI toys: ‘I’m always listening’
Addictive, AI-enabled tablets are taking over, and also, Woody is balding in the new Toy Story movie, out June 19.
AI’s promise to indie filmmakers: Faster, cheaper, lonelier
AI expands access to filmmaking for resource-constrained creators. But as efficiency becomes the industry’s north star, creativity risks being overwhelmed by a deluge of low-effort, AI-generated content.
General Catalyst commits $5B to India over five years
The pledge marks a sharp jump from General Catalyst's earlier $500 million–$1 billion India earmark.
Every Little Helps: Building Knowledge Graph Foundation Model with Fine-grained Transferable Multi-modal Tokens
arXiv:2602.15896v1 Announce Type: new Abstract: Multi-modal knowledge graph reasoning (MMKGR) aims to predict the missing links by exploiting both graph structure information and multi-modal entity contents. Most existing works are designed for a transductive setting, which learns dataset-specific embeddings and...
Missing-by-Design: Certifiable Modality Deletion for Revocable Multimodal Sentiment Analysis
arXiv:2602.16144v1 Announce Type: new Abstract: As multimodal systems increasingly process sensitive personal data, the ability to selectively revoke specific data modalities has become a critical requirement for privacy compliance and user autonomy. We present Missing-by-Design (MBD), a unified framework for...
Beyond Learning: A Training-Free Alternative to Model Adaptation
arXiv:2602.16189v1 Announce Type: new Abstract: Despite the continuous research and evolution of language models, they sometimes underperform previous versions. Existing approaches to overcome these challenges are resource-intensive, highlighting the need for alternatives that enable immediate action. We assume that each...
MemoryArena: Benchmarking Agent Memory in Interdependent Multi-Session Agentic Tasks
arXiv:2602.16313v1 Announce Type: new Abstract: Existing evaluations of agents with memory typically assess memorization and action in isolation. One class of benchmarks evaluates memorization by testing recall of past conversations or text but fails to capture how memory is used...
B-DENSE: Branching For Dense Ensemble Network Learning
arXiv:2602.15971v1 Announce Type: new Abstract: Inspired by non-equilibrium thermodynamics, diffusion models have achieved state-of-the-art performance in generative modeling. However, their iterative sampling nature results in high inference latency. While recent distillation techniques accelerate sampling, they discard intermediate trajectory steps. This...
Fast Online Learning with Gaussian Prior-Driven Hierarchical Unimodal Thompson Sampling
arXiv:2602.15972v1 Announce Type: new Abstract: We study a type of Multi-Armed Bandit (MAB) problems in which arms with a Gaussian reward feedback are clustered. Such an arm setting finds applications in many real-world problems, for example, mmWave communications and portfolio...