AI Frontlines

The Intersection of Artificial Intelligence and Modern Warfare

by @GregMushen and @KyleShannon

The Big Stuff

AI Warfare

"Amateurs talk about tactics, but professionals study logistics," goes a popular military adage, emphasizing that wars are often won and lost on the foundation of logistics. This sentiment was echoed by Antoine-Henri Jomini, a renowned French general and military strategist, who said, "Logistics is the practical art of moving armies." Now, with the advent of Large Language Models (LLMs) designed for military operations and coordination, the mastery of logistics, once the exclusive domain of seasoned military professionals, is on the verge of being commoditized.

Palantir, a leading data analytics and AI firm, released a demo of their new product AIP, a large language model to help with military logistics and strategy. The model is a chatbot for military operators that aggregates data from federated sources of information, both public and private.

Here is a summary of the product:

1. Notifications - AIP will alert operators of different things, such as anomalous enemy activity

2. Surveillance On-Demand - Operators can input surveillance tasks, such as "send a drone out to get more data", and then see that drone response in the UI in realtime

3. Attack Recommendations - From there, operators can ask for attack recommendations, then send those up the chain of command for approval

4. Attack Routing and Planning - Once approved, AIP can determine which nearby unit should be sent, determine how many munitions they will need, determine the optimal route, and plan things like signal jamming, "assign jammers to each of the validated high priority communications targets"

Palantir is utilizing multiple open source large language models, which we have talked about for the last two weeks. Specifically, they are using GPT-NeoX-20B fine tuned of course specifically for this purpose, along with Google's open source Flan-T5-XL, and Dolly-v2-12b, which we wrote about two weeks ago.

Think 2016 disinformation was bad? AI, "Hold my beer."

In March, Donald Trump first shared an AI-generated image of him praying. We are now starting to see the first 2024 political attack ads, using AI-generated imagery, in this case, of President Biden. We can expect the 2024 political season to be filled with AI-generated imagery, videos, memes, and content.

Companies like Midjourney, the maker's of the popular image generation software by the same name, are grappling with how to deal with this. They said in a recent office hours that they may need to change the rules around public figures and clamp down a bit prior to the political season. They said they are trying to strike the balance between people's expressive freedom, and using the platform solely for sensationalism. They have admitted they don't know how to deal with this, and they are open to suggestions. Will this lead to arguments of political bias in image generation platforms?

Microsoft's AI Business Results

Microsoft had an earnings call this week. Their earnings results highlight the financial impact of AI products. Azure OpenAI Service customers are up 10x QoQ, over 10k organizations have signed up for Copilot for Business, Teams "intelligent recaps" has thousands of paid customers just two months in, Bind has more than 100M daily active users, and search revenue grew 10%. (Hat tip to @nonmayorpete for an amazing summary).

Other Big News

• Cruise, a self-driving company, is now operating 24/7 in all of San Francisco, and announces rollout in other cities soon (link)

• OpenAI now allows users to turn off their chat history (link)

• Runway launches app to bring users video-to-video generative AI (link)

• NVIDIA releases NeMo Guardrails, allowing enterprises to keep applications safe (link)

• Hugging Face releases HuggingChat (link)

• Grimes announces that she will allow her AI voice to be used with 50% royalties (link)

• TikTok releases an AI Avatar creator (link)

• Replit releases replit-code-v1-3b, a code generation model that is 40% better than comparable models (link)

• Apple is reportedly building a paid AI health coach (link)

• Microsoft releases Microsoft Designer, an AI-powered tool to help create compelling designs (link)

• Meta mentions AI 27x in earnings call, alludes to AI chat and image creation tools (link)

Smaller But Still Cool Things:

• I became addicted to running, and lost 26 lbs. All thanks to ChatGPT (link)

• I decided to outsource my entire personal financial life to GPT-4 (link)

• AI Hits - a collection of AI-generated music (link)

• PricewaterhouseCoopers to Pour $1 Billion Into Generative AI (link)

• Will AI change online shopping? (link)

Going Deeper

• Andrew NG releases a free prompt engineering course (link)

• Scaling Transformers to 1M tokens and beyond with RMT (link)

• Are you using LLMs to scrape websites? If so, check out Scrapeghost, an experimental library for scraping websites without writing page-specific code (link)

• PMC-Patients, a large-scale dataset of patient summaries is now available on Hugging Face (link)

• Answering Questions by Meta-Reasoning over Multiple Chains of Thought (link)

• A Cookbook of Self-Supervised Learning (link)

Tweets of the Week

Eye Candy

Do you have 30 seconds for a quick survey to help us improve Everday AI?

We'd love your feedback! Click here.

Do you like what you're reading? Share it with a friend.