Trustworthiness of generative language models (GLMs) is crucial in their deployment to critical decision making systems.

In Deep Reinforcement Learning (RL), it is a challenge to learn representations that do not exhibit catastrophic forgetting or interference in non-stationary environments.

To this end, we employ a pre-trained vision LLM [20] as the interface and propose a novel Adapter-in-Adapter (AnA) framework, allowing efficient extension and alignment for other modalities.

When do mesa-optimization algorithms emerge in autoregressively trained transformers?

Intelligent education stands as a prominent application of machine learning.