·
DataBubble
  • Home
  • Models
  • News
  • Compare
  • Boards
  • Pricing
  • About
  • Newsletter
  • Methodology
  • Contact
Latest
The AI graveyard: a running list of projects and startups that didn’t make it2h◆Asian governments react to Nvidia CEO Jensen Huang’s exit3h◆The OpenAI board member who fired Sam Altman says we still can’t trust AI CEOs5h◆Jensen Huang and the decline of the visionary showman founder CEO8h◆Mach2: Scientific Discovery Platform with Deterministic Execution Constraints15h◆OverServe: Resource Amplification Attacks on SaaS Agents via Tool--Prompt Co-Design15h◆CFGComposer: Compositional Audio-Text-to-Motion Generation via Copula Score Correction15h◆Multi-View 3D Reconstruction Using V-JEPA Features15h◆GingivitisSeg: A Public Dataset and Comparative Benchmark for Gingivitis Segmentation in Intraoral Images15h◆Can Model Answer Probabilities Represent Survey Populations? Conditional Prediction and Joint Validity in Jev Survey Simulation15h◆Contrastive Bootstrap Representation Learning for Label-Efficient Agricultural Image Classification15h◆SO-Arena: Benchmarking the Capability Spectrum and Reliability of System One Models15h◆The BEATS Dataset: Towards Understanding Music-Physiology Dynamics in Real-World Scenarios15h◆RESYNC: Residual Latent-State Synchronization for Deadline-Bound Streaming Inference15h◆LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks15h◆Do More Modalities Always Help? A Geometric Perspective on Missing-Modality Robustness15h◆Which Buildings Are Artificial Intelligence-Ready? A Measurement-Based Assessment Framework for AI Question Answering and Actuation15h◆Catch-Only-One: Non-Transferable Examples for Model-Specific Authorization15h◆DAISY: Fault-Guided Assertion Inference for Dafny15h◆Individualized Algorithmic Advice as a Strategic Signal on Competitive Markets15h◆The AI graveyard: a running list of projects and startups that didn’t make it2h◆Asian governments react to Nvidia CEO Jensen Huang’s exit3h◆The OpenAI board member who fired Sam Altman says we still can’t trust AI CEOs5h◆Jensen Huang and the decline of the visionary showman founder CEO8h◆Mach2: Scientific Discovery Platform with Deterministic Execution Constraints15h◆OverServe: Resource Amplification Attacks on SaaS Agents via Tool--Prompt Co-Design15h◆CFGComposer: Compositional Audio-Text-to-Motion Generation via Copula Score Correction15h◆Multi-View 3D Reconstruction Using V-JEPA Features15h◆GingivitisSeg: A Public Dataset and Comparative Benchmark for Gingivitis Segmentation in Intraoral Images15h◆Can Model Answer Probabilities Represent Survey Populations? Conditional Prediction and Joint Validity in Jev Survey Simulation15h◆Contrastive Bootstrap Representation Learning for Label-Efficient Agricultural Image Classification15h◆SO-Arena: Benchmarking the Capability Spectrum and Reliability of System One Models15h◆The BEATS Dataset: Towards Understanding Music-Physiology Dynamics in Real-World Scenarios15h◆RESYNC: Residual Latent-State Synchronization for Deadline-Bound Streaming Inference15h◆LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks15h◆Do More Modalities Always Help? A Geometric Perspective on Missing-Modality Robustness15h◆Which Buildings Are Artificial Intelligence-Ready? A Measurement-Based Assessment Framework for AI Question Answering and Actuation15h◆Catch-Only-One: Non-Transferable Examples for Model-Specific Authorization15h◆DAISY: Fault-Guided Assertion Inference for Dafny15h◆Individualized Algorithmic Advice as a Strategic Signal on Competitive Markets15h◆
News/Safety as a Budget: Calibrating the Control Channel of Constrained GRPO for Vision-Language Models
arxiv
PublishedNovember 14, 2026 at 5:00 AM

Safety as a Budget: Calibrating the Control Channel of Constrained GRPO for Vision-Language Models

Source
arxiv.orgfull article ↗
Read on arxiv→
Publisher summary· verbatim

arXiv:2611.16628v1 Announce Type: new Abstract: Reinforcement learning can teach vision-language models (VLMs) to decline harmful image-text requests, but how safe the result becomes is set indirectly, through reward weights found by trial runs. Stating safety as budgets on the rate of harmful assis

Stay posted· Newsletter

Three short emails a week: Monday, Wednesday and Friday. Top movers, new models and the most-referenced story.

// no spam · unsubscribe one-click · free forever

Your email is used only to send The Bubble Brief. Privacy policy

Discussion
Source
↗
arxiv
Read original ↗All from arxiv →

No replies yet. Be first.

Source
↗
arxiv
Read original ↗All from arxiv →

Related coverage

More from ARXIV
arxivMach2: Scientific Discovery Platform with Deterministic Execution Constraints15harxivOverServe: Resource Amplification Attacks on SaaS Agents via Tool--Prompt Co-Design15harxivCFGComposer: Compositional Audio-Text-to-Motion Generation via Copula Score Correction15harxivMulti-View 3D Reconstruction Using V-JEPA Features15h
The Bubble Brief
MON · WED · FRI

Read AI insights every Monday, Wednesday and Friday — top movers, new models, story of the week.

// no spam · unsubscribe one-click · free forever

Your email is used only to send The Bubble Brief. Privacy policy

Originally published on arxiv ↗
PrivacyTermsRefundsLegal notice|AboutMethodologyContact

© 2026 Databubble

Built by Marouane Gazouzi · Readyo, AI interview coach

HomeModelsNews