This step is seen as a reflection of growing concerns about the ethical and security dimensions of artificial intelligence.
Last year, when OpenAI’s ChatGPT sparked worldwide interest, many powerful figures in both Silicon Valley and Washington DC were taken aback. Now, the US government is expected to have advance notice of future artificial intelligence (AI) developments, such as large language models, the technology behind ChatGPT. The Biden administration is preparing to use the Defense Production Act to force tech companies to notify the government when they train an AI model using significant amounts of computing power. This rule could take effect next week.
- 104 million dollars in sales in 2 minutes! Honor Magic 6 series introduced
- Man’s space adventure began 67 years ago
The new requirement will give the US government access to critical information about the most sensitive projects of OpenAI, Google, Amazon and other tech companies competing in the AI space. The companies will also have to provide information about the security tests conducted on new AI creations.
OpenAI has remained tight-lipped about how much work is being done on the successor to its current flagship product, GPT-4. The US government may be the first to know that work or security testing on GPT-5 has actually begun. OpenAI did not immediately respond to a request for comment.
The US has set interesting rules for artificial intelligence developers! You can’t do anything without our knowledge!
US Secretary of Commerce Gina Raimondo said, “We are using the Defense Production Act, the authority we have thanks to the President. It requires companies to notify us every time they train a new big language model and share the results – the security data – so we can scrutinize it.” He made this statement at an event at Stanford University’s Hoover Institution. He did not say when it would take effect or what action the government would take on information it receives about AI projects. More details are expected to be announced next week.
The new rules are being implemented under a sweeping White House executive order issued last October. The executive order gave the Commerce Department until January 28 to issue a regulation under which it would have to inform US authorities about powerful new AI models that are emerging. These details must include the amount of computing power used, ownership of the data fed into the model, and security testing details, the order said.
The October order wants to begin defining the time at which AI models require reporting to the Department of Commerce, but initially sets a level set at 100 septillion (one million billion billion or more) floating-point operations per second, or flops, and 1,000 times lower for large language models running on DNA sequence data. Neither OpenAI nor Google has disclosed the amount of computing power they used to train their most powerful models GPT-4 and Gemini, but a congressional research service report on the executive order suggests it was slightly more than was used to train GPT-4.
Raimondo also confirmed that they will soon implement another requirement of the October executive order. This requirement requires cloud computing providers such as Amazon, Microsoft and Google to inform the government when a foreign company uses their resources to train a large language model. Foreign projects must be reported when they pass the same initial threshold of 100 septillion flops.
