Making AI models ‘forget’ undesirable data hurts their performance

So-called “unlearning” strategies are used to make a generative AI mannequin overlook particular and undesirable data it picked up from coaching knowledge, like delicate non-public knowledge or copyrighted materials. However present unlearning strategies are a double-edged sword: They may make a mannequin like OpenAI’s GPT-4o or Meta’s Llama 3.1 405B a lot much less able to answering […]

© 2024 TechCrunch. All rights reserved. For private use solely.