To use Ask Sage, the Army had access to 100 million tokens as part of an annual “Enterprise Pack” subscription. A token represents a unit of text or image output from LLM. For the Ask Sage tool, one token is equivalent to approximately 3.7 characters, according to documents seen by WIRED. According to Breaking Defense, the Pentagon burned around 20 billion tokens per day during Operation Epic Fury over 38 days in Iran.
The Army and the Department of Defense did not respond to requests for comment. Ask Sage was no different.
It is unclear whether the tokens used by regular Department of Defense employees are drawn from the same pool as employees who may be using AI tools on classified or classified information. Still, the Department of Defense’s emphasis on AI continues. The Intercept reported Monday that the Pentagon continues to rely on AI tools and has reduced staffing at the Civilian Protection Center of Excellence, which is tasked with preventing civilian casualties in conflict zones. Instead, the Pentagon is developing AI tools to speed up the assessments normally performed by center staff.
The Army is not the first company eager to reimagine the near-unlimited use of generative AI. After encouraging employees to “tokenmaxx,” Meta quietly removed the leaderboards that tracked token usage and is now working to curb usage. Last week, Meta’s head of Instagram, Adam Mosseri, announced the idea of capping the amount of tokens used by each of the company’s engineers. According to a report from Fortune, Uber engineers used up a year’s worth of generated AI tokens in just four months.
The Army official said he didn’t find generative AI tools particularly useful for their work, and that when he did use them, he found them to be unreliable. One model even claimed to have completed a task that was not completed. “I think there’s definitely some aspects of the U.S. federal bureaucracy where these tools could be useful. However, thoughtless application and use will not lead to an effective, efficient, or reliable deployment.”
This article originally appeared on Wired.com.
