·
DataBubble
  • Home
  • Models
  • News
  • Compare
  • Boards
  • Pricing
  • About
  • Newsletter
  • Methodology
  • Contact
Latest
Google’s AI search is rapidly becoming the default, new data shows43m◆Power up your AI infrastructure! A first look at the Smart Systems Stage agenda at TechCrunch Disrupt 20261h◆This $9 key physically locks your most addictive apps1h◆Ilya Sutskever’s Safe Superintelligence partners with Nvidia to scale its AI research1h◆Enigma raises $70M to make controlling a robot as easy as adjusting the volume3h◆Nvidia, Microsoft launch open AI security alliance — without OpenAI, Google, or Anthropic4h◆The path to artificial superintelligence4h◆Closing the data loop in AI-driven drug discovery5h◆Building the enterprise environment for agentic AI5h◆NVIDIA Cosmos-H-Dreams: Bringing Real-Time Generative Simulation to Surgical Robotics7h◆A Consensus-Based Framework for Relative Preference Evaluation of Large Language Models12h◆Probing Latent Colombian Identity Inferences in Qwen2.5-7B with Natural Language Autoencoders12h◆Data Quality over Capacity: Internalizing Documents into LoRA Adapters for Closed-Book QA12h◆Enjoy Your Talk: A Human-Centered Benchmark for Multi-Turn Dialogue with Decoupled User Simulation, Target Modeling, and Judging12h◆Multi-Mask Diffusion Language Models for Few-Step Generation12h◆Solar Open 2 Technical Report12h◆The Geometry of Personality: Activation Steering with Jungian Cognitive Functions12h◆Self-Guided Process Reward Optimization with Redefined Step-wise Advantage for Process Reinforcement Learning12h◆H$^2$SD: Hybrid Hindsight Self-Distillation12h◆LunarFM: A Shared Multimodal Representation of the Moon's Surface12h◆Google’s AI search is rapidly becoming the default, new data shows43m◆Power up your AI infrastructure! A first look at the Smart Systems Stage agenda at TechCrunch Disrupt 20261h◆This $9 key physically locks your most addictive apps1h◆Ilya Sutskever’s Safe Superintelligence partners with Nvidia to scale its AI research1h◆Enigma raises $70M to make controlling a robot as easy as adjusting the volume3h◆Nvidia, Microsoft launch open AI security alliance — without OpenAI, Google, or Anthropic4h◆The path to artificial superintelligence4h◆Closing the data loop in AI-driven drug discovery5h◆Building the enterprise environment for agentic AI5h◆NVIDIA Cosmos-H-Dreams: Bringing Real-Time Generative Simulation to Surgical Robotics7h◆A Consensus-Based Framework for Relative Preference Evaluation of Large Language Models12h◆Probing Latent Colombian Identity Inferences in Qwen2.5-7B with Natural Language Autoencoders12h◆Data Quality over Capacity: Internalizing Documents into LoRA Adapters for Closed-Book QA12h◆Enjoy Your Talk: A Human-Centered Benchmark for Multi-Turn Dialogue with Decoupled User Simulation, Target Modeling, and Judging12h◆Multi-Mask Diffusion Language Models for Few-Step Generation12h◆Solar Open 2 Technical Report12h◆The Geometry of Personality: Activation Steering with Jungian Cognitive Functions12h◆Self-Guided Process Reward Optimization with Redefined Step-wise Advantage for Process Reinforcement Learning12h◆H$^2$SD: Hybrid Hindsight Self-Distillation12h◆LunarFM: A Shared Multimodal Representation of the Moon's Surface12h◆
News/Do Active SAE Feature Planes Carry More Holonomy? A Preregistered Reversal in Gemma
arxiv
PublishedJuly 24, 2026 at 4:00 AM
—neutral

Do Active SAE Feature Planes Carry More Holonomy? A Preregistered Reversal in Gemma

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

arXiv:2607.20522v1 Announce Type: new Abstract: This paper tests whether holonomy concentrates on active sparse-autoencoder (SAE) feature planes in Gemma 2 2B, a concrete operationalization of the broader semantic-concentration prediction. Holonomy is measured at the final-token layer-12 to layer-13

Stay posted· Newsletter

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

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

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
arxivA Consensus-Based Framework for Relative Preference Evaluation of Large Language Models12harxivProbing Latent Colombian Identity Inferences in Qwen2.5-7B with Natural Language Autoencoders12harxivData Quality over Capacity: Internalizing Documents into LoRA Adapters for Closed-Book QA12harxivEnjoy Your Talk: A Human-Centered Benchmark for Multi-Turn Dialogue with Decoupled User Simulation, Target Modeling, and Judging12h
The Bubble Brief
WEEKLY

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

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

Originally published on arxiv ↗
HomeModelsNews