The MEND method corrects and updates large AI models

New method for updating large AI models

Language models, which are large AI networks, can contain incorrect or outdated information. MEND demonstrates how to update LLMs while not changing the entire network.

Many AI applications such as image analysis, natural language processing and image generation use large AI models. OpenAI’s GPT-3 is an example of a model that has more capabilities than smaller, specialized models. It can also be fine-tuned to improve its performance.

Even the biggest AI models make mistakes, and they also contain outdated data. GPT-3 has data that dates back to 2019, when Theresa May still held the position of prime minister.