Lucas Kupssinskü

New GUARD-IT method unlearns LLMs at inference time without parameter changes

Researchers have developed a new method called GUARD-IT for inference-time machine unlearning, which aims to remove specific data's influence from large language models without altering their parameters. This technique …