Google is using large language models (LLMs) to flag abuse of its products to enforcement teams. The company says that it can now build and train a model in a matter of days, instead of weeks or months, to find specific kinds of abuse.
This is especially valuable for new and emerging abuse areas, as Google can quickly prototype a model that is an expert in finding a specific type of abuse and automatically route it to its teams for enforcement.
The specific LLMs used by Google remain undisclosed, the company’s Senior Director of Trust and Safety, Amanda Storey, shared insights at the Fighting Misinformation Online Summit in Brussels on October 26, 2023. She emphasized Google’s commitment to balancing access to information with user safety, underlining the responsibility to provide trustworthy content.
Google’s strategy revolves around protecting users from harm, delivering reliable information, and collaborating with experts and organizations to ensure a safer online environment. Their decision-making process is guided by principles that prioritize user diversity, personal choice, and the freedom of expression while mitigating harmful content’s proliferation.
To achieve this, Google is continually evolving its tools, policies, and techniques, with a strong focus on harnessing artificial intelligence (AI) for abuse detection. Notably, they have developed a prototype using LLMs, which can rapidly identify and address content abuse at scale. This innovative approach shows promising results in proactively protecting users from emerging risks.
In addition to using LLMs to fight misinformation, Google is also taking other steps to reduce the threat and promote trustworthy information in its generative AI products. These steps include launching new tools, adapting its policies, and partnering with others.
Google is also partnering with other organizations to fight misinformation. For example, the company has committed $10 million to the Global Fact Checking Fund and has partnered with think tanks, civil society organizations, and fact-checking networks to combat misinformation about the war in Ukraine.
The sources for this piece include an article in TheVerge.