Emanuel Ruzak

on the atlas — 29

• Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations6 savers
• Announcing the ARC White-Box Estimation Challenge — LessWrong1 savers
• Discovering Backdoor Triggers — LessWrong1 savers
• sapientinc/HRM-Text: HRM-Text is a 1B text generation model based on the HRM architecture, strengthened by task completion and latent space reasoning. ·1 savers
• p(doom)1 savers
• Airtable - MATS 9.0 Research Poster Gallery2 savers
• Vanessa Kosoy's Shortform — LessWrong1 savers
• Recursive forecasting: Eliciting long-term forecasts from myopic fitness-seekers — AI Alignment Forum1 savers
• Jo Jiao, Boyd Kane - Limited-parameter finetuning [MATS Symposium | Winter 2026] - YouTube1 savers
• Joan Velja, Lennie Wells - Building an empirical Science of AI Debate [MATS Symposium | Winter 2026] - YouTube1 savers
• [2302.05442] Scaling Vision Transformers to 22 Billion Parameters1 savers
• Automation collapse — LessWrong1 savers
• Prospects for Alignment Automation: Interpretability Case Study — LessWrong1 savers
• [2604.16812] Introspection Adapters: Training LLMs to Report Their Learned Behaviors4 savers
• Encyclical Letter of His Holiness Leo XIV Magnifica Humanitas (15 May 2026)45 savers
• LoRA Without Regret - Thinking Machines Lab32 savers
• Introducing talkie: a 13B vintage language model from 193017 savers
• Dario Amodei — Policy on the AI Exponential16 savers
• Automated Weak-to-Strong Researcher15 savers
• Inkling: Our Open-Weights Model - Thinking Machines Lab11 savers
• Recommendations for Technical AI Safety Research Directions7 savers
• Physical Intelligence (π)7 savers
• Modifying LLM Beliefs with Synthetic Document Finetuning6 savers
• The Lesson to Unlearn6 savers
• NOUS RESEARCH – The AI accelerator company4 savers
• Interpreting Language Model Parameters4 savers
• Optimal Timing for Superintelligence3 savers
• Working through a small tiling result — LessWrong3 savers
• Tensor-Transformer Variants are Surprisingly Performant — LessWrong3 savers