Let’s Talk About AI Safety, Computer Science

Draws parallels between AI ‘scheming’ and ape language experiments, exploring deceptive tendencies in LLMs and the need for advanced monitoring for AI safety.

Evaluates the ability of frontier LLMs to persuade users on harmful topics, assessing their strategies and the implications for AI safety and ethics.

Investigates how LLMs can be tuned to become more susceptible to jailbreaking, highlighting the implications for AI safety and the need for robust defenses.

Examines the challenges of predicting AI agent behavior from observed actions and its implications for AI safety, alignment, and the need for robust monitoring.

A diary entry covering AI personalities, utility engineering for emergent value systems, and methods for evaluating the goal-directedness of Large Language Models (LLMs).

A diary entry on AI risks, including misalignment, misuse, and s-risks, and an exploration of emergent misalignment due to prompt sensitivity in LLMs.

A diary entry on longtermism and its moral implications for the future, and a paper on teaching models to verbalize reward hacking in Chain-of-Thought reasoning.

A diary entry on common use cases for AI models and the risks of models obfuscating their reasoning to evade safety monitors.

A diary entry on advanced prompt engineering techniques and the faithfulness of LLM self-explanations for commonsense tasks.

A diary entry on how Chain-of-Thought (CoT) reasoning affects LLM’s ability to evade monitors, and the challenge of unfaithful reasoning in model explanations.