Understanding HTML with Large Language Models
Explores the capabilities of large language models (LLMs) for HTML understanding and contributes HTML understanding models and an in-depth analysis of their capabilities under three tasks.
LLMs pretrained on standard natural language corpora transfer remarkably well to HTML understanding tasks, creating and open-sourcing a large-scale HTML dataset distilled and auto-labeled from CommonCrawl can promote further research on LLMs for HTML understanding.
In-Context Policy Iteration
Presents In-Context Policy Iteration (ICPI), an algorithm for performing Reinforcement Learning (RL) in-context, using foundation models without expert demonstrations or gradients
ICPI is a promising algorithm for RL tasks in low-resource settings, eliminating the role of in-weights learning, which can be slow and sacrificing the "few-shot" quality of in-context learning.