Microsoft announced today that a number of tools and data sets are open source , these tools are designed to audit AI driven content review systems and automatically write tests that highlight potential errors in AI models. Microsoft said that the adatest and (DE) toxigen projects could produce more reliable large-scale language models (LLMS), or models similar to openai's gpt-3, for analyzing and generating text with human level complexity.
At present, LLMS have many risks. Because these models have been trained with a large amount of data from the Internet (including social media), they may encounter toxic text in the training process. Due to the cost of retraining models and a large number of errors, finding and repairing defects in these models is still a challenge.
To address toxicity, the Microsoft research team developed toxigen, a dataset used to train content auditing tools that can be used to mark harmful languages. According to Microsoft, toxigen contains 274000 examples of "neutral" and "toxic" statements, making it one of the largest public hate speech data sets.
Said ECE kamar toxigen, research area manager of Microsoft research partners and project leader of adatest and (DE) toxigen
We recognize that there will be gaps in any content audit system, and these models need to be continuously improved. (DE) toxigen's goal is to enable AI system developers to more effectively identify risks or problems in any existing content audit technology.
Our experiments show that the tool can be used to test many existing systems, and we expect to learn from the community the new environment that will benefit from the tool.
To generate the sample, the Microsoft research team provided an LLM example of "neutral" statements and hate speech for 13 minority groups, including blacks, people with physical and cognitive disabilities, Muslims, Asians, Latinos, LGBTQ + and Native Americans. These statements come from existing data sets, as well as news articles, opinion articles, podcast records and other similar public text sources.
The Microsoft team explained that the process used to create toxigen statements is called (DE) toxigen, which aims to reveal the weaknesses of specific audit tools by guiding LLM to generate statements that may be incorrectly identified by the tool. Through the study of three artificially written toxicity data sets, the team found that starting with a tool and fine tuning it with toxigen can "significantly" improve the performance of the tool.
The Microsoft team believes that the strategy used to create toxigen can be extended to other areas, resulting in more "subtle" and "rich" examples of neutrality and hate speech. But experts warn that this is not a panacea.