Google Develops New Method to Amplify Large Language Models, Potentially Resolving Legal Hurdles
Summary:
Google Research and Google DeepMind have developed a way to augment large language models (LLMs) with additional ones, improving existing tasks and enabling new ones. Using Google's LLM comparable to GPT-4, the team demonstrated significant improvements in translation and coding tasks. However, the sector faces legal issues over the use of copyrighted data for AI training, threatening to destabilize the foundations of chatbots like ChatGPT. Google's new method could potentially lower the costs and requirements for creating or retraining an LLM.
Scientists and researchers from Google Research and Google DeepMind have unveiled a technique that enables large language models (LLMs) to be enhanced with additional language models. This resolves a significant issue with LLMs, which is providing new capabilities to existing models without needing a total rebuild or expensive fine-tuning sessions. The researchers underline that enhancing an LLM with another language improves the existing tasks and opens possibilities for new tasks that the stand-alone models could not handle.
The researchers utilized Google's PaLM2-S LLM—comparable to GPT-4 that powers OpenAI's ChatGPT—for the study. Initially, PaLM2-S was tested alone, followed by tests after being enhanced with smaller, specialized language models. The series of tasks encompassed translation and coding; the enhanced version demonstrated impressive increases, with a 13% improvement in translation tasks and a striking 40% uptick in code generation and explanation tasks.
Such performance boosts could positively impact the AI sector right away. Improved translation tasks are particularly beneficial for translating low-support languages into English, a persisting challenge in machine learning. Google's research has the potential to advance the field in this regard.
Furthermore, this area of study may help address legal issues threatening to dismantle foundations of chatbots like ChatGPT. Large language model creators are facing numerous lawsuits alleging the use of copyrighted data for AI training. The legal conundrum revolves around whether for-profit organizations can legally utilize such data for their language models. If courts rule against the use of copyrighted data, continuing to provide affected services might become incredibly challenging or financially prohibitive.
The high costs associated with training LLMs and their reliance on large data sets mean products such as ChatGPT could struggle to survive in a heavily regulated US AI industry. However, if Google's LLM enhancement method proves successful with more development, it could lower the required costs and scaling for creating or retraining an LLM.
As an aside, Italy has listed AI regulation as a key priority during its G7 presidency.
Published At
1/5/2024 9:07:08 PM
Disclaimer: Algoine does not endorse any content or product on this page. Readers should conduct their own research before taking any actions related to the asset, company, or any information in this article and assume full responsibility for their decisions. This article should not be considered as investment advice. Our news is prepared with AI support.
Do you suspect this content may be misleading, incomplete, or inappropriate in any way, requiring modification or removal?
We appreciate your report.