ARMY INTELLIGENCE
Doyle’s team is working to transition CamoGPT into a program of record and put more focus into developing an Army- specific LLM. “An Army-specific LLM would perform much better at recognizing Army acronyms and providing recommen- dations founded in Army doctrine,” Doyle said. “Our team does not train LLMs; we simply host published, open-source models that have been trained by companies like Meta, Google and Mistral.”
TECH-FOCUSED
Work is increasingly completed using computers and digital apps, as demonstrated by Staff Sgt. Andrew Schede, 105th Communications Squadron cyber systems operations, in the communications building at Stewart Air National Guard Base, New York, on March 6, 2025. (Photo by Senior Airman Sarah Post, 105th Airlift Wing)
One of those organizations is the U.S. Army, which now has an Army-specific chatbot known as CamoGPT. Currently boasting 75,000 users, CamoGPT started development in the fall of 2023 and was first deployed in the spring of 2024. Live data is important for the Army and other companies that choose to implement their own AI chat-based applications. CamoGPT does not currently have access to the internet because it is still in a proto- type phase, but connecting to the net is a goal. Another goal for the platform is to accurately respond to questions that involve current statistics and high-stakes information. What’s more, CamoGPT can process classified data on SIPRNet and unclassified information on NIPRNet.
THE MORE THE MERRIER Large language models (LLMs) are a type of AI chatbot that can understand and generate human language based on
62 Army AL&T Magazine Summer 2025
inputs. LLMs undergo extensive train- ing, require copious amounts of data and can be tedious to create. Tey can also process and respond to information as a human would. Initially, the information fed to the bot must be input individually and manually by human beings until a pattern is established, at which point the computer can take over. Updating facts can be a daunting task when considering the breadth of data from around the world that AI is expected to process.
Aidan Doyle, a data engineer at the Army Artificial Intelligence Integration Center (AI2C), works on a team of three active- duty service members, including another data engineer and a data analyst, as well as four contracted software developers and one contracted technical team lead. “It’s a small team, roles are fluid, [and] every- one contributes code to Camo[GPT],” Doyle said.
Te process of training LLMs involves pre- training the model by showing it as many examples of natural language as possi- ble from across the internet. Everything from greetings to colloquialisms must be input so it can mimic human conversation. Sometimes, supervised learning is neces- sary for specific information during the fine-tuning step. Ten the model gener- ates different answers to questions, and humans evaluate and annotate the model responses and flag problems that arise. Once preferred responses are identified, developers adjust the model accordingly. Tis is a post-training step called rein- forcement learning with human feedback, or alignment. Finally, the model generates both the questions and answers itself in the self-play step. When the model is ready, it is deployed.
A LITTLE TOO CREATIVE Te use of AI in creative fields faces signif- icant challenges, such as the potential for plagiarism and inaccuracy. Artists spend a lot of time creating work that can easily be duplicated by AI without giving the artist any credit. AI can also repackage copy- righted material. It can be tough to track down the original source for something when it is generated with AI.
AI can—and sometimes does—intro- duce inaccuracies. When AI fabricates information unintentionally, it is called a hallucination. AI can make a connection
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68 |
Page 69 |
Page 70 |
Page 71 |
Page 72 |
Page 73 |
Page 74 |
Page 75 |
Page 76 |
Page 77 |
Page 78 |
Page 79 |
Page 80 |
Page 81 |
Page 82 |
Page 83 |
Page 84 |
Page 85 |
Page 86 |
Page 87 |
Page 88 |
Page 89 |
Page 90 |
Page 91 |
Page 92 |
Page 93 |
Page 94 |
Page 95 |
Page 96 |
Page 97 |
Page 98 |
Page 99 |
Page 100
