·
DataBubble
  • Home
  • Models
  • News
  • Compare
  • Boards
  • Pricing
  • About
  • Newsletter
  • Methodology
  • Contact
Latest
AI researchers stranded Claude on a virtual desert island. It started to “play.”2h◆RCPN: towards robust pseudo-labels in unsupervised CLIP adaptation via consistent prototypes and neighborhoods8h◆Can Graph Learning Learn Circuits?8h◆Beyond Fixed Acquisition Functions: Temporal Difference Active Learning for Perturbation Discovery8h◆Trajectory-Guided Steering: Mitigating Spurious Feature Bias in Clinical Classification8h◆Sentence-Level Topic Modeling with Lexical and Contextual Evidence8h◆Referring Multi-Object Tracking with Diverse Dynamic Expressions8h◆R$^2$ Flow: Recursive Self-Improvement via Recursive Skill Evolution8h◆Logic-Guided Graph Denoising Diffusion Models8h◆Unit-level aggregation breaks coverage in causal foundation models8h◆Mask-Based Deep Symbolic Regression8h◆From Events to Reasoning: Hierarchical Rule Composition over Explicit States8h◆ECHOES: Seeing Beyond the Current Image via Recursive History Compression and Privileged On-Policy Distillation8h◆Sensitivity Analysis of Decision Rules in an Agent-Based Agro-Ecological Model8h◆PRIME: Towards Verifiable Evidence-Grounded Reasoning in LLM Agents8h◆From Natural Language Intent to Executable Control: Hierarchical Feedback-Guided LLM Planning for Multi-component Cyber-Physical Systems8h◆Functional Welfare for AI Safety8h◆FRACTAL: A Self-Similar Multi-Agent Orchestration Architecture with Formal Communication, Convergence, and Depth Bounds8h◆Repairing LLM-Generated Specifications with Deterministic Diagnosis8h◆At the Limits of Computability: Steps Towards Autonomous AI Discovery on the Busy Beaver Frontier8h◆AI researchers stranded Claude on a virtual desert island. It started to “play.”2h◆RCPN: towards robust pseudo-labels in unsupervised CLIP adaptation via consistent prototypes and neighborhoods8h◆Can Graph Learning Learn Circuits?8h◆Beyond Fixed Acquisition Functions: Temporal Difference Active Learning for Perturbation Discovery8h◆Trajectory-Guided Steering: Mitigating Spurious Feature Bias in Clinical Classification8h◆Sentence-Level Topic Modeling with Lexical and Contextual Evidence8h◆Referring Multi-Object Tracking with Diverse Dynamic Expressions8h◆R$^2$ Flow: Recursive Self-Improvement via Recursive Skill Evolution8h◆Logic-Guided Graph Denoising Diffusion Models8h◆Unit-level aggregation breaks coverage in causal foundation models8h◆Mask-Based Deep Symbolic Regression8h◆From Events to Reasoning: Hierarchical Rule Composition over Explicit States8h◆ECHOES: Seeing Beyond the Current Image via Recursive History Compression and Privileged On-Policy Distillation8h◆Sensitivity Analysis of Decision Rules in an Agent-Based Agro-Ecological Model8h◆PRIME: Towards Verifiable Evidence-Grounded Reasoning in LLM Agents8h◆From Natural Language Intent to Executable Control: Hierarchical Feedback-Guided LLM Planning for Multi-component Cyber-Physical Systems8h◆Functional Welfare for AI Safety8h◆FRACTAL: A Self-Similar Multi-Agent Orchestration Architecture with Formal Communication, Convergence, and Depth Bounds8h◆Repairing LLM-Generated Specifications with Deterministic Diagnosis8h◆At the Limits of Computability: Steps Towards Autonomous AI Discovery on the Busy Beaver Frontier8h◆
News/The Last Fingerprint: How Markdown Training Shapes LLM Prose
arxiv
PublishedApril 1, 2026 at 4:00 AM
—neutral

The Last Fingerprint: How Markdown Training Shapes LLM Prose

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

arXiv:2603.27006v1 Announce Type: cross Abstract: Large language models produce em dashes at varying rates, and the observation that some models "overuse" them has become one of the most widely discussed markers of AI-generated text. Yet no mechanistic account of this pattern exists, and the paralle

Models mentioned
02
  • 01meta-llama logo
    Llama
    meta-llama/Llama
  • 02openai logo
    gpt-4
    openai/gpt-4
    0.0%IN $30.00/Mtok
Compare these 2 models→
Related
05
  • arxivJul 10
    A Vision Toward Energy-Efficient Domain-Specific Artificial Intelligence Models and Agents
  • arxivJul 2
    WorkBench Revisited: Workplace Agents Two Years On
  • arxivJun 25
    Representation Interventions Enable Lifelong Knowledge Memory Control in LLMs
  • arxivMay 19
    EmoMind: Decoding Affective Captions from Human Brain fMRI
  • arxivMay 11
    End-to-end PDDL Planning with Hardcoded and Dynamic Agents
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
02
  • 01
    Llama
    meta-llama/Llama
  • 02
    gpt-4
    openai/gpt-4
Source
↗
arxiv
Read original ↗All from arxiv →
Tags
04
#language models#training data#fine-tuning#markdown
Mentioned companies
05
AnthropicOpenAIMetaGoogleDeepSeek

No replies yet. Be first.

Mentioned models
02
  • 01
    Llama
    meta-llama/Llama
  • 02
    gpt-4
    openai/gpt-4
Source
↗
arxiv
Read original ↗All from arxiv →
Tags
04
#language models#training data#fine-tuning#markdown
Mentioned companies
05
AnthropicOpenAIMetaGoogleDeepSeek

Related coverage

More from ARXIV
arxivRCPN: towards robust pseudo-labels in unsupervised CLIP adaptation via consistent prototypes and neighborhoods8harxivCan Graph Learning Learn Circuits?8harxivBeyond Fixed Acquisition Functions: Temporal Difference Active Learning for Perturbation Discovery8harxivTrajectory-Guided Steering: Mitigating Spurious Feature Bias in Clinical Classification8h
The Bubble Brief
MON · WED · FRI

Read language models 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