There is no denying that the arrival of ChatGPT through OpenAI marks the first and last in how the world knows and relates to artificial intelligence. Currently, the United States government has focused its sights on technology, which, by law, is required to provide advance notice to the US government of future AI developments that involve good language models.
The Biden-Harris administration intends that the use of the Defense Production Act can force technology companies to report when training an AI model. According to specialized sources such as Wired, the rule could be implemented before the end of January.
Under this scheme, the government of the United States has access to key information on sensitive projects of technologies such as Google, Amazon, OpenAI, and others experimenting with AI using large amounts of computing power. In this sense, if the use is approved, the organizations must provide information for the security tests created by your AI creations.
The United States confirmed its intention to lead the field of artificial intelligence after publishing an executive order in October 2023. This order gives the Department of Commerce until January 28, 2024, to develop a plan to achieve the purpose of compiling companies to provide details and data about new and powerful AI models that are still in development. Such data should include computer-generated information about ownership of the data entered into the model and details of security testing.
“We used the Defense Production Act, which is an authority we have thanks to the president, to do a study that requires companies to tell us every time they train a new model in multiple languages and share with us those results, the safety data. , so we can review them,” said Gina Raimondo, Secretary of Commerce of the United States, on January 26, 2024, at an event at the Hoover Institution at Stanford University.
Raimondo did not disclose: When will this new requirement be implemented. nor what steps the government can take regarding information from AI projects. More details are expected to be released in February 2024.
He also confirmed that the Department of Commerce will soon implement another requirement in the October executive order. This requires cloud computing providers, such as Amazon, Microsoft, and Google, to notify the government if a foreign company is using its resources to train a large language model. Foreign projects must be notified if they exceed the same initial threshold of 100 octillion failures.