·
DataBubble
  • Home
  • Models
  • News
  • Compare
  • Boards
  • Pricing
  • About
  • Newsletter
  • Methodology
  • Contact
Latest
SFMambaNet: Spectral-Frequency Enhanced Selective State Space Model for Correspondence Pruning3h◆Optical-Guided Neural Collapse for SAR Few-Shot Class Incremental Learning3h◆Dynamic Infilling Anchors for Format-Constrained Generation in Diffusion Large Language Models3h◆Temporal Order Matters for Agentic Memory: Segment Trees for Long-Horizon Agents3h◆Why Muon Outperforms Adam: A Curvature Perspective3h◆Provably Auditable and Safe LLM Agents from Human-Authored Ontologies3h◆Efficient Reasoning on the Edge3h◆Treat Traffic Like Trees: A Semantic-Preserving Hierarchical Graph-Based Expert Framework for Encrypted Traffic Analysis3h◆Rollout-Level Advantage-Prioritized Experience Replay for GRPO3h◆ANN Search: Recall What Matters3h◆Self-Evolving Deep Research via Joint Generation and Evaluation3h◆Smart Picks in the Dark: Towards Efficient RLVR for Reasoning via Tracing Metacognitive Pivots3h◆GeoMin: Data-Efficient Semi-Supervised RLVR via Geometric Distribution Modeling3h◆Aligning Deep Implicit Preferences by Learning to Reason Defensively3h◆SMAC-Talk: A Natural Language Extension of the StarCraft Multi-Agent Challenge for Large Language Models3h◆Consensus is Strategically Insufficient: Reasoning-Trace Disagreement as a Knowledge-Representation Signal3h◆VAMPS: Visual-Assisted Mathematical Problem Solving Benchmark3h◆StepPRM-RTL: Stepwise Process-Reward Guided LLM Fine-Tuning for Enhanced RTL Synthesis3h◆Talk is (Not) Cheap: A Taxonomy and Benchmark Coverage Audit for LLM Attacks3h◆FedMental: Evaluating Federated Learning for Mental Health Detection from Social Media Data3h◆SFMambaNet: Spectral-Frequency Enhanced Selective State Space Model for Correspondence Pruning3h◆Optical-Guided Neural Collapse for SAR Few-Shot Class Incremental Learning3h◆Dynamic Infilling Anchors for Format-Constrained Generation in Diffusion Large Language Models3h◆Temporal Order Matters for Agentic Memory: Segment Trees for Long-Horizon Agents3h◆Why Muon Outperforms Adam: A Curvature Perspective3h◆Provably Auditable and Safe LLM Agents from Human-Authored Ontologies3h◆Efficient Reasoning on the Edge3h◆Treat Traffic Like Trees: A Semantic-Preserving Hierarchical Graph-Based Expert Framework for Encrypted Traffic Analysis3h◆Rollout-Level Advantage-Prioritized Experience Replay for GRPO3h◆ANN Search: Recall What Matters3h◆Self-Evolving Deep Research via Joint Generation and Evaluation3h◆Smart Picks in the Dark: Towards Efficient RLVR for Reasoning via Tracing Metacognitive Pivots3h◆GeoMin: Data-Efficient Semi-Supervised RLVR via Geometric Distribution Modeling3h◆Aligning Deep Implicit Preferences by Learning to Reason Defensively3h◆SMAC-Talk: A Natural Language Extension of the StarCraft Multi-Agent Challenge for Large Language Models3h◆Consensus is Strategically Insufficient: Reasoning-Trace Disagreement as a Knowledge-Representation Signal3h◆VAMPS: Visual-Assisted Mathematical Problem Solving Benchmark3h◆StepPRM-RTL: Stepwise Process-Reward Guided LLM Fine-Tuning for Enhanced RTL Synthesis3h◆Talk is (Not) Cheap: A Taxonomy and Benchmark Coverage Audit for LLM Attacks3h◆FedMental: Evaluating Federated Learning for Mental Health Detection from Social Media Data3h◆
News/Prompt-Induced Score Variance in Zero-Shot Binary Vision-Language Safety Classification
arxiv
PublishedMay 4, 2026 at 4:00 AM
—neutral

Prompt-Induced Score Variance in Zero-Shot Binary Vision-Language Safety Classification

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

arXiv:2605.00326v1 Announce Type: new Abstract: Single-prompt first-token probabilities from zero-shot vision-language model (VLM) safety classifiers are treated as decision scores, but we show they are unreliable under semantically equivalent prompt reformulation: even when the binary label is cons

Stay posted· Newsletter

A 5-min weekly brief — top movers, price watch, story of the week.

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

Discussion
Mentioned models
01
  • 01
    zero-shot vision-language model
Source
↗
arxiv
Read original ↗All from arxiv →
Tags
04
#safety#benchmark#calibration#reliability

No replies yet. Be first.

Mentioned models
01
  • 01
    zero-shot vision-language model
Source
↗
arxiv
Read original ↗All from arxiv →
Tags
04
#safety#benchmark#calibration#reliability

Related coverage

More from ARXIV
arxivSFMambaNet: Spectral-Frequency Enhanced Selective State Space Model for Correspondence Pruning3harxivOptical-Guided Neural Collapse for SAR Few-Shot Class Incremental Learning3harxivDynamic Infilling Anchors for Format-Constrained Generation in Diffusion Large Language Models3harxivTemporal Order Matters for Agentic Memory: Segment Trees for Long-Horizon Agents3h
The Bubble Brief
WEEKLY

Read safety insights every Tuesday — top movers, new releases, story of the week.

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

Originally published on arxiv ↗
HomeModelsNews