policy

PolicyCliff

SafeAGI-01 · PyTorch

or hover any field below to flag it

Overview

Name
PolicyCliff
Author
SafeAGI-01
Framework
PyTorch
License
unknown
Skill type
navigation
Evidence level
untested
Task description
The Policy Cliff: A Theoretical Analysis of Reward-Policy Maps in Large Language Models. A rigorous mathematical framework analyzing the stability of the reward–policy mapping in RL-trained LLMs.

Spaces

Action space
other · 0-dim · 0Hz
Observation space
  • type: other

Links

HuggingFace repo
null
Paper (arXiv)
null

Compatible robots

20

Compatible environments

0

No environments list PolicyCliff yet.

Datasets that reference this policy

0

No datasets reference PolicyCliff yet.