alt Hacker NewsDoes anyone have any idea what the DoD could possibly want from OpenAI? Less accurate/more sycophantic missiles?
Replies
Some of the more popular models (NIPRGPT, the various DREN models) are “soft banned” and DoD is in need of a unified solution. MSFT’s GCC HIGH and GovCloud implementations have been slow to materialize. But more to your point - everyone is using LLM’s to pick up the slack from layoffs. Im sitting in meetings and watching my gov customers generate documentation and proposals everyday. Everything the commercial world uses AI for the US gov is doing the same. Cant directly speak to targeting but you can bet your ass there are 100 different offensive projects trying to integrate AI into ISR work and the like.
Yeah, tons. SIGNT / HUMINT analysis. After action report summaries. war gaming to optimize deterrence. human machine teaming. LLM-in-the-loop for warfighters. rapid code gen in field deployments for units to spin up software solutions. The list is endless, imho.
Automatically generated, native sounding, propaganda at scale - capable of interacting in real time. This was always the MIC money endgame for LLMs. This is also probably why they are enlisting tech execs from Meta, OpenAI, etc.
No. People see "defense" and immediately think everything is a weapon, for some reason. The DoD is still an organization with exactly the same business functions as wherever you work. This is likely going to be something more or less like the ChatGPT for Gov thing. You ask it to summarize videos and articles, review a document you're writing, spruce it up, whatever the rest of you are doing with LLMs, but you're allowed to use it from a locked down DoD workstation that can't access public web services.
AI explosives with personalities feature in https://en.m.wikipedia.org/wiki/Dark_Star_(film)
You will be surprise how much work at the DoD has nothing to do with weapons.
> “This contract, with a $200 million ceiling, will bring OpenAI’s industry-leading expertise to help the Defense Department identify and prototype how frontier AI can transform its administrative operations, from improving how service members and their families get health care, to streamlining how they look at program and acquisition data, to supporting proactive cyber defense,”
Translated - they'll hand out GPT access to a bunch of service members and administrators. Except the UI will have a big DoD logo and words like "SECURE" and "CLASSIFIED" will be displayed on it a few dozen times.
ChatGPT, do you know where the General left his keys?
You realize that the DoD has a huge amount of normal business work like logistics, project management, people management, benefits management, etc? Right?
Not that the bomb answers: "I am sorry Dave, i can't do that!"
I would guess it’s for mass surveillance. Even just the ability to extract names and entities from audio, video, and text on every piece of public media would be useful.
An on premise deployment ?
Easy PT plans
Sycophantic missiles would be desirable
To attempt to make sense of all my after action reports? kidding
Maybe one potential use could be to drink from the firehose of data and then try to create summary bullet-point reports for the higher ups instead of relying on data filtering up the chain of command in the old game of telephone. At least that is what I would use if for. Getting that data in nearly real-time in an accurate presentation would be priceless.
I do not have the slightest idea how they will secure all this data if going to a 3rd party like OpenAI unless they have their own self hosted version of it on their own mainframes. If that data is going to live in a 3rd party they need to secure their systems in a magical way systems has never been secured. They would have to cast some seriously powerful protection spells.
1. Secretary of Defense feels like bombing some place. Asks aide to write a report on, justification, logistics, and consequences.
2. Aide tells subordinate to write report.
3. Subordinate uses ChatGPT to write the 100-page report. Sends it to aide.
4. Aide uses ChatGPT to summarize report. Sends summary to SecDef.
5. SecDef accidentally posts summary on publicly-accessible social media page, then forwards to President.
6. Bombs go boom.