Palantir has already sold domestic surveillance services to the U.S. Immigration and Customs Enforcement Service, so it’s no surprise that the company founded by billionaire Peter Thiel is also working on a Pentagon foray. On Tuesday, the company released a video demo of its latest product, his Palantir Artificial Intelligence Platform (AIP). The system itself is simply designed to integrate large-scale language models (LLMs) like OpenAI’s GPT-4 and Google’s BERT into privately-operated networks, but the first thing they did was It was to apply it to the modern battlefield.
In the video demo above, a military operator tasked with monitoring the Eastern European theater of war spotted enemy forces massing near the border and called on a ChatGPT-style digital assistant to help deploy reconnaissance drones. response, initiate tactical responses to perceived attacks, and even organize jamming of enemy communications. AIP responds to an operator’s request for better photographs and on reconnaissance missions he fires Reaper drones to estimate the enemy’s composition and capabilities by proposing an appropriate response given the discovery of armored elements. It has been shown to help
“LLMs and algorithms must be controlled in this highly regulated and sensitive context to ensure they are used in a lawful and ethical manner,” the video begins. To that end, AIP’s operations are based on three “key pillars,” he said. First, AIP can be deployed across sensitive systems to enable real-time analysis of both sensitive and non-sensitive data for ethical and legal purposes. The company doesn’t elaborate on how it works. The second pillar will allow users to switch scopes and actions for all her LLMs and assets on the network. According to the demo, AIP itself generates a secure digital record of the entire operation. This is “important for mitigating significant legal, regulatory and ethical risks in sensitive sensitive settings”. AIP’s “industry-leading guardrail” for
Palantir’s scenario has a “human in the loop” to prevent such behavior, but from the video it looks like the “operator” just nods along with what the AIP suggests. . The demo also doesn’t detail what steps are taken to prevent the LLMs that the system relies on from “hallucinating” relevant facts and details.
