HaoYang Shang

New approach improves stance prediction by addressing LLM training failures

Researchers have identified four failure modes in applying test-time scaling and post-training techniques to individual stance prediction tasks. These failures include incorrect consensus, selection errors, response ove…