Robert Kirk

Deep Ignorance: Filtering Pretraining Data Builds Tamper-Resistant Safeguards into Open-Weight LLMs

Reward Model Overoptimisation in Iterated RLHF

Dataset Featurization: Uncovering Natural Language Features through Unsupervised Data Reconstruction

Existing Adversarial LLM Unlearning Evaluations Are Inconclusive

Poisoning Attacks on LLMs Require a Near-constant Number of Poison Samples

How Do Large Language Monkeys Get Their Power (Laws)?

Investigating Non-Transitivity in LLM-as-a-Judge

Fundamental Limitations in Pointwise Defences of LLM Finetuning APIs

Procedural Knowledge in Pretraining Drives Reasoning in Large Language Models