·
DataBubble
  • Home
  • Models
  • News
  • Compare
  • Boards
  • Pricing
  • About
  • Newsletter
  • Methodology
  • Contact
Latest
Meta removes language that raised scraping and AI training concerns18h◆Anthropic says its workforce has become less junior and is spending more time in meetings19h◆Replit makes OpenAI its default over Anthropic, CEO says21h◆Meta removes language that may have allowed it to scrape news sites1d◆Websites are closing their doors to AI agents1d◆How Itaú is building generative AI for its entire customer base1d◆Roundtables: A Conversation With the Creator of AI-Designed Viruses1d◆India becomes the latest country to report rogue AI agent activity1d◆Physical AI is more than robots, says the Nvidia veteran now investing in it1d◆An Nvidia insider-turned-VC on why robotics hasn’t hit its ChatGPT moment, yet1d◆Elorian’s first AI model wants to reason with images, not just about them1d◆Jev is Vercel’s fastest-adopted AI model ever2d◆A week with RTX Spark: everything we’ve learned using Nvidia’s new laptops2d◆Mark Wahlberg on using AI in film ‘the right way’ and being open to an AI clone carrying on his legacy2d◆Meta’s new terms may allow it to train AI on news sites for free2d◆Veritas expands its anti-counterfeit chip to jewelry, trophies, and fragrance bottles2d◆Replit’s CEO says the future of work is the ‘self-driving company’2d◆Optimal Hydra: Model Heterogeneous Federated Learning for LLMs using Data-Informed Hydra-Nets2d◆Judging the Judges: Confidence and Disagreement in Rubrics as Rewards2d◆Content associations with answer selection and engagement in supply chain community question answering2d◆Meta removes language that raised scraping and AI training concerns18h◆Anthropic says its workforce has become less junior and is spending more time in meetings19h◆Replit makes OpenAI its default over Anthropic, CEO says21h◆Meta removes language that may have allowed it to scrape news sites1d◆Websites are closing their doors to AI agents1d◆How Itaú is building generative AI for its entire customer base1d◆Roundtables: A Conversation With the Creator of AI-Designed Viruses1d◆India becomes the latest country to report rogue AI agent activity1d◆Physical AI is more than robots, says the Nvidia veteran now investing in it1d◆An Nvidia insider-turned-VC on why robotics hasn’t hit its ChatGPT moment, yet1d◆Elorian’s first AI model wants to reason with images, not just about them1d◆Jev is Vercel’s fastest-adopted AI model ever2d◆A week with RTX Spark: everything we’ve learned using Nvidia’s new laptops2d◆Mark Wahlberg on using AI in film ‘the right way’ and being open to an AI clone carrying on his legacy2d◆Meta’s new terms may allow it to train AI on news sites for free2d◆Veritas expands its anti-counterfeit chip to jewelry, trophies, and fragrance bottles2d◆Replit’s CEO says the future of work is the ‘self-driving company’2d◆Optimal Hydra: Model Heterogeneous Federated Learning for LLMs using Data-Informed Hydra-Nets2d◆Judging the Judges: Confidence and Disagreement in Rubrics as Rewards2d◆Content associations with answer selection and engagement in supply chain community question answering2d◆
News/Measuring Evaluation-Context Divergence in Open-Weight LLMs: A Paired-Prompt Protocol with Pilot Evidence of Alignment-Pipeline-Specific Heterogeneity
arxiv
PublishedMay 8, 2026 at 4:00 AM
—neutral

Measuring Evaluation-Context Divergence in Open-Weight LLMs: A Paired-Prompt Protocol with Pilot Evidence of Alignment-Pipeline-Specific Heterogeneity

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

arXiv:2605.06327v1 Announce Type: cross Abstract: Safety benchmarks are routinely treated as evidence about how a language model will behave once deployed, but this inference is fragile if behavior depends on whether a prompt looks like an evaluation. We define evaluation-context divergence as an ob

Models mentioned
02
  • 01meta-llama logo
    Llama-3.1-8B
    meta-llama/Llama-3.1-8B
    DL 652K-1.2%IN $0.10/Mtok
  • 02meta-llama logo
    Llama-3.1-70B
    meta-llama/Llama-3.1-70B
Compare these 2 models→
Related
05
  • arxivJul 29
    FusionML: Prefill, Not Decode - Mechanism and Boundaries of CPU+GPU Co-Execution on Unified-Memory Apple Silicon
  • arxivJul 18
    Branching Policy Optimization: Sandbox-Native Language Agent Reinforcement Learning
  • arxivJul 16
    A Shared Subcircuit Lets LLMs Count Down Across Tasks
  • techcrunchJun 30
    Vibe-coding platform Base44 launches own model as AI startups seek defensibility
  • arxivMay 28
    A Benchmark Construction and Evaluation Framework for Specialist Domains: Case Study on Defense-related Documents
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
Mentioned models
07
  • 01
    OLMo-3
  • 02
    OLMo-3-Instruct
  • 03
    Mistral-Small-3.2
  • 04
    Phi-3.5-mini
  • 05
    Llama-3.1-8B
    meta-llama/Llama-3.1-8B
    0.7M dl
  • 06
    Llama-3.1-70B
    meta-llama/Llama-3.1-70B
  • 07
    Llama-Guard-3-8B
Source
↗
arxiv
Read original ↗All from arxiv →
Tags
04
#safety#benchmark#evaluation#language models

No replies yet. Be first.

Mentioned models
07
  • 01
    OLMo-3
  • 02
    OLMo-3-Instruct
  • 03
    Mistral-Small-3.2
  • 04
    Phi-3.5-mini
  • 05
    Llama-3.1-8B
    meta-llama/Llama-3.1-8B
    0.7M dl
  • 06
    Llama-3.1-70B
    meta-llama/Llama-3.1-70B
  • 07
    Llama-Guard-3-8B
Source
↗
arxiv
Read original ↗All from arxiv →
Tags
04
#safety#benchmark#evaluation#language models

Related coverage

More from ARXIV
arxivOptimal Hydra: Model Heterogeneous Federated Learning for LLMs using Data-Informed Hydra-Nets2darxivJudging the Judges: Confidence and Disagreement in Rubrics as Rewards2darxivContent associations with answer selection and engagement in supply chain community question answering2d
The Bubble Brief
MON · WED · FRI

Read safety 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