Sagnik Chatterjee - Reading List

Reading list

What I am reading lately. 6 of 65 read in 2026.

To be read

December 2026 (2)

Emergent Misalignment: Narrow Finetuning can Produce broadly misaligned LLMs - 690S
07 December 2026
Universal and Transferable Adverserial Attacks on Aligned Language Models - 690S
02 December 2026

November 2026 (15)

Differential Privacy Has Disparate Impact on Model Accuracy - 684
29 November 2026
Spinning Language Models - Risks of Propoganda-As-A-Service and Countermeasures - 684
23 November 2026
OverThink: SlowDown Attacks on Reasoning LLMS - 684
22 November 2026
Mechanistic Interpretability for AI Safety : A Review - 690S
18 November 2026
Defeating Prompt Injections by Design - 684
17 November 2026
Contextual Agent Security - A Policy for Every Purpose - 684
17 November 2026
Measuring Progress on Scalable Oversight for Large Language Models - 690S
16 November 2026
Emergent Misalignment: Narrow Finetuning can produce broadly misaligned LLMs - 684
15 November 2026
OpenAI- HuggingFace Incident - 690S
11 November 2026
Unsolved Problems in ML Safety - 690S
09 November 2026
SoK: Watermarking for AI-Generated Content - 684
08 November 2026
Cooperative Inverse Reinforcement Learning - 690S
04 November 2026
How to Backdoor Federated Learning - 684
03 November 2026
Are Aligned Neural Networks Adverserially Aligned? -684
01 November 2026
Self-Interpreting Adversarial Images - 684
01 November 2026

October 2026 (12)

Eureka: Human Level Reward Design via Coding Large Language Models - 690S
28 October 2026
Safe Imitation Learning via Fast Bayesian Reward Inference from Preferences - 690S
26 October 2026
Universal and Transferable Adversarial Attacks on Aligned Language Models - 684
25 October 2026
Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection - 684
25 October 2026
Direct Preference Optimization: Your Language Model is Secretely a Reward Model - 690S
21 October 2026
Models of Human Preference for learning reward functions - 690S
19 October 2026
Extrapolating Beyong Suboptimal Demonstrations via Inverse Reinforcement Learning from Observations - 690S
14 October 2026
Generative Adverserial Imitation Learning - 690S
06 October 2026
Can LLMS Keep a Secret ? Testing Privacy Implications of Language Models via Contextual Integrity Theory - 684
06 October 2026
AirGapAgent: Protecting Privacy-Conscious Conversational Agents - 684
06 October 2026
Beyond Memorization: Violating Privacy Via Inference with Large Language Models - 684
04 October 2026
Can Large Language Models Really Recognize Your Name? - 684
04 October 2026

September 2026 (27)

Efficient Probabilisitc Performance Bounds for Inverse Reinforcement Learning - 690S
30 September 2026
Maximum Entropy Inverse Reinforcement Learning - 690S
29 September 2026
Differentially Private Synthetic Data via Foundation Model APIs 1: Images - 684
29 September 2026
Learning Differentially Private Recurrent Language Models - 684
29 September 2026
Deep Learning with Differential Privacy - 684
27 September 2026
Scaling Laws for Differentially Private Language Models - 684
27 September 2026
Apprenticeship Learning via Inverse Reinforcement Learning -690S
23 September 2026
Mechanistic Interpretability for AI Safety: A Review.
22 September 2026
A Mathematical Framework for Transformer Circuits.
22 September 2026
Interpretability in the Wild: A Circuit for Indirect Object Identification in GPT-2 Small.
22 September 2026
Towards Best Practices of Activation Patching in Language Models: Metrics and Methods.
22 September 2026
Toy Models of Superposition.
22 September 2026
Sparse Autoencoders Find Highly Interpretable Features in Language Models
22 September 2026
Position: Causality is Key for Interpretability Claims to Generalise
22 September 2026
The Linear Representation Hypothesis and the Geometry of Large Language Models.
22 September 2026
Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models.
22 September 2026
Refusal in Language Models Is Mediated by a Single Direction.
22 September 2026
Open Problems in Mechanistic Interpretability.
22 September 2026
Communication-Efficient Learning of Deep Networks from Decentralized Data- 684
22 September 2026
Advances and Open Problems in Federated Learning -684
22 September 2026
Extracting Training Data from Large Language Models - 684
20 September 2026
Language Models May Verbatim Complete Text They Were Not Explictly Trained On - 684
20 September 2026
Membership Inference Attacks Cannot Prove that a Model Was Trained On Your Data - 684
20 September 2026
Imitation Attacks and Defenses for Black-Box Machine Translation System - 684
20 September 2026
Membership Inference Attacks from First Principles - 684
15 September 2026
Membership Inference Attacks against Machine Learning Models - 684
15 September 2026
Towards the Science of Security and Privacy in Machine Learning - 684
13 September 2026

August 2026 (3)

What is the purpose of Interpretability?
31 August 2026
What is so hard about continual learning?
30 August 2026
Reinforcement Learning for LLMS
25 August 2026

Completed

September 2026 (5)

Specification Gaming: Flip Side of AI Ingenuity - 690S
21 September 2026
Goal Misgeneralization: Why Correct Specifications Aren't Enough for Correct Goals - 690S
21 September 2026
RL Theory for CS690S ,Part1 , Part2 and Proximal Policy Optimization
16 September 2026
A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning - 690S
10 September 2026
Behaviour Cloning - 690S
08 September 2026

August 2026 (1)

Current AI seems pretty misaligned
30 August 2026