11 papers
Pessimism's Paradox: Conservative Offline Training Amplifies Reward Hacking During Online Adaptation in Reasoning Models
Subramanyam Sahoo, Aman Chadha, Vinija Jain +1
Conservative offline training is widely advocated as a safe foundation for subsequent online adaptation: if a policy stays close to well-supported behaviour, the argument goes, it…
Linear Probes Detect Task Format, Not Reasoning Mode in Language Model Hidden States
Subramanyam Sahoo, Vinija Jain, Aman Chadha +1
Linear probing of large language model (LLM) hidden states is widely used to claim that models learn distinct representations for different reasoning types. We test this by probing…
The Reasoning Trap -- Logical Reasoning as a Mechanistic Pathway to Situational Awareness
Subramanyam Sahoo, Aman Chadha, Vinija Jain +1
Situational awareness, the capacity of an AI system to recognize its own nature, understand its training and deployment context, and reason strategically about its circumstances, i…
SAHOO: Safeguarded Alignment for High-Order Optimization Objectives in Recursive Self-Improvement
Subramanyam Sahoo, Aman Chadha, Vinija Jain +1
Recursive self-improvement is moving from theory to practice: modern systems can critique, revise, and evaluate their own outputs, yet iterative self-modification risks subtle alig…
When Shallow Wins: Silent Failures and the Depth-Accuracy Paradox in Latent Reasoning
Subramanyam Sahoo, Aman Chadha, Vinija Jain +1
Mathematical reasoning models are widely deployed in education, automated tutoring, and decision support systems despite exhibiting fundamental computational instabilities. We demo…
I Can't Believe It's Not Robust: Catastrophic Collapse of Safety Classifiers under Embedding Drift
Subramanyam Sahoo, Vinija Jain, Divya Chaudhary +1
Instruction tuned reasoning models are increasingly deployed with safety classifiers trained on frozen embeddings, assuming representation stability across model updates. We system…