As a Trump According to procurement material reviewed by Intercept, management and its executive executives are trying to rebuild the administrative state.
With Elon Musk’s so-called government efficiency office set up in the IRS, there is a wider drive to replace the federal bureaucracy with machine learning software. Art NVIDIA SuperPod AI Computing Cluster. According to a previously unreported February 5 acquisition document, the setup will combine 31 individual NVIDIA servers. Each includes eight company flagship blackwell processors designed to train and operate electric tools such as ChatGPT.
The hardware has not been purchased and installed yet. Prices are not listed, but the SuperPod system is reportedly starting at $7 million. Please note that the setup described in the contract documentation includes a substantial memory upgrade from NVIDIA.
Though small compared to the large AI training data centers deployed by companies like Openai and Meta, the SuperPod is a powerful and expensive setup using cutting-edge technology provided by NVIDIA. Although hardware can be used in a variety of ways, it is sold as a turnkey means of creating and queriing AI models. Last year, Miter Corporation, a federal government’s R&D lab, acquired a $20 million superpod setup to train custom-built AI models for use by agencies, and a massive increase in US computing power. “The purchase was promoted as “.
It is unclear how the IRS will use the Superpods exactly. A spokesperson for the agency said the IRS had no information to share about purchasing a supercomputer. A 2024 report by the generals of the Ministry of Tax and Treasury identified 68 different AI-related projects ongoing in the IRS. Nvidia clusters are not named within them, but many have been edited.
However, some clues can be collected from the purchase materials. “The IRS requires a robust and scalable infrastructure that can handle complex machine learning (ML) workloads,” the document explains. “The NVIDIA Superpod is a critical component of this infrastructure and provides the computational power, storage and networking capabilities needed to support the development and deployment of large-scale ML models.”
This document notes that superpods are performed by the IRS research, applied analysis, statistical department, or RAAS. Although no specific uses have been cited, the Compliance Data Warehouse Project for this department is behind this superpod purchase, and the Compliance Data Warehouse Project for this department was previously automated fraud detection, identity theft They use machine learning to prevent it, and generally states that they “have a deeper understanding of the mechanisms that drive it.” Taxpayer behavior. ”
“The IRS has more unique data than most institutions that are probably completely undeveloped.”
It is unclear from the documents whether the purchase of the Superpod was planned under the Biden administration or whether it represents a new initiative from the Trump administration.
Travis Thompson, a tax attorney at Boutin Jones with expertise in IRS AI strategies, said funding from the 2022 Inflation Reduction Act was generally allocated to upgrade IRS technology. But “The IRS is heading towards AI well before IRA funding,” Thompson explained. “They didn’t have enough money to properly enforce the tax law. They were looking for ways to do more with less.” Government Accountability Bureau, June 2024 The report suggested that the IRS uses artificial intelligence-based software to “potentially lack hundreds of millions of dollars (that) each year.”
Thompson added that the agency is ripe for machine learning training because of the mountains of personal and financial data at the top. “The IRS probably have more unique data than most institutions that are completely undeveloped. Looking at things like this Nvidia cluster and training machine learning algorithms in the future, it’s completely logical because there’s data. It’s a good thing. AI needs data. It needs a lot. And it needs it quickly. And the IRS has it.”
Purchases will be made at the intersection for US governance of artificial intelligence technology. During Trump’s first term, the RAAS office was assigned “responsibility to oversee and oversee AI in the IRS.” This is based on Executive Order 13960, which I signed just before I left the office in 2020. “Implementing AI by the US – an approach that is no longer preferred by the American Technology Baron, which currently advocates for the fierce development of these technologies in consideration of ethics and risks. One of Trump’s first moves after he took office was to reverse the Biden administration’s executive order for a bigger AI safety guardrail in government use.
Many of the AI industry, which are “safe AI,” have become close ally of the new Trump White House, including Elon Musk and venture capitalist Mark Andreesen. This wing of Silicon Valley reportedly encouraged the new administration to use artificial intelligence to help dismantle administrative states through automation.
This week, the Wall Street Journal reported that mask liquidators had arrived at the IRS. It has long reported that the target of disparity and distortion by Trump and Republican allies is an institution. A few days ago, the New York Times stated that “information technology information for tax collectors, with the aim of representing so-called departments of government efficiency, is to automate more work to replace the need for human staff members. I’ve been asking for it.”
While the IRS has been increasingly turning to AI for automated fraud detection and chatbot-based support services, including collaboration with NVIDIA, the new NVIDIA supercomputer is reducing agency personnel whenever possible It could also benefit people who are interested in this. . The Washington Post’s February 8 report cited an unnamed federal official who described masks as “replacing the human labor force with machines,” saying, “It’s all we can automate machinery.” “He said. And technocrats replace bureaucrats. ”
Musk’s Underling is reportedly thinking about replacing the Ministry of Education people with a large language-based chatbot.
Wired previously reported that Musk loyalty Thomas Shedd, who was the director within General Services Administration, spoke about Trump’s second-generation “AI-First” agenda. Doge staff are reportedly already relying on Microsoft’s Azure AI platform to seek advice on how to significantly reduce programs. The Nvidia SuperPod could not be done with its own replication service, such as Microsoft’s offering, but it is powerful enough to train AI models based on government data.
Thompson told Intercept that his efforts to cut the federal workforce and acquire more aggressive hands that fit artificial intelligence systems.
“I firmly believe that it is rooted behind the reduction in the human workforce, which is what appears to be the current management goal, where it is comprehensive to implement more technology-based systems. “There are goals,” he explained. “If you’re going to reduce your workforce, something has to pick up sagging. Something has to do the job.”
