Dylan Sam

Personal Website

Publications 2026 When Should We Introduce Safety Interventions During Pretraining? Dylan Sam, Sachin Goyal, Pratyush Maini, Alexander Robey, J. Zico Kolter Under Review [pdf] 2025 Safety Pretraining: Toward the Next Generation of Safe AI Pratyush Maini*, Sachin Goyal*, Dylan Sam*, Alex Robey, Yash Savani, Yiding Jiang, Andy Zou, Zachary C. Lipton, J. Zico Kolter NeurIPS, 2025 [pdf, website] Predicting the Performance of Black-box LLMs through Follow-up Queries Dylan Sam, Marc Finzi, and J. Zico Kolter NeurIPS, 2025 ICML Reliable and Responsible Foundation Models, 2025 [pdf,…

saved by

• Ishaan Panigrahi

related reading

• When Should We Introduce Safety Interventions During Pretraining?arxiv.org
• Synthetic Persona Pretraining: Alignment from Token Zeromodelraising.ai
• Conditional misalignment: common interventions can hide emergent misalignment behind contextual triggersarxiv.org
• [2601.10160] Alignment Pretraining: AI Discourse Causes Self-Fulfilling (Mis)alignmentarxiv.org
• SFT Drives Gemini’s Safety Properties — LessWronglesswrong.com
• ARENA - AI Safety Curriculumlearn.arena.education
• gpt-4.pdfcdn.openai.com
• AI in 2025: gestalt — LessWronglesswrong.com
• Student Projects - CS 2881R AI Safetyboazbk.github.io
• Alignment Pretraining: AI Discourse Causes Self-Fulfilling (Mis)alignment — LessWronglesswrong.com
• [2604.22082] Removing Sandbagging in LLMs by Training with Weak Supervisionarxiv.org
• Alignment Pretraining: AI Discourse Causes Self-Fulfilling (Mis)alignmentalignmentpretraining.ai