Generative AI Could Redefine How Countries Coordinate Their Militaries

Welcome to the 8th issue of the AstroFeather AI newsletter!

It's been another big week for AI! Google just wrapped up its 2023 I/O event, and spoiler alert, they're integrating generative AI into many of their products; Palantir's AIP system could one day be used to orchestrate militaries on the battlefield; I found some promising AI projects that could impact the fashion world and the way we clean our homes; and Wendy's and Google are training AI to take your next order at the drive-thru.

I hope you enjoy reading this week’s updates and if you have any helpful feedback, feel free to respond to this email or contact me directly [ LinkedIn ]!

Thanks - Adides Williams, Founder @ AstroFeather

[ Daily AI News | Newsletter Signup | Past Newsletters ]

In today’s recap (10 min read time):

• How Generative AI (GenAI) Could Redefine the Battlefield.

• (Research) AI Systems that Can Generate Fashion Videos, Clean Rooms, and Learn from Multiple Sources.

• Product Launches and Updates from Google, IBM, and Anthropic.

• Company Announcements from Wendy’s, Anthropic, Google, and Amazon.

Must-Read News Articles and Updates

Update #1. How Generative AI (GenAI) Could Redefine the Battlefield.

The Latest: Palantir Technologies' Artificial Intelligence Platform, AIP, is a chat-driven system that could one day be used to coordinate military units in real time, and the "unprecedented demand" it's currently experiencing adds to ongoing discussions about the wartime use of AI and autonomous weapons systems (AWS) - systems that can identify, lock onto, and attack a target without human intervention.

How Palantir AIP works: AIP uses large language models (Dolly-v2-12b, Flan-T5XL, and GPT-NeoX-20b) to respond to questions from "military operators" about battle conditions, including enemy position and weapon capabilities, and requests for a recommended course of action (COA). During a recent demonstration, a military operator responded to a fictitious combat scenario in which she was responsible for monitoring activity in Eastern Europe. With AIP, the military operator was able to use text prompts to accomplish the following:

• Quickly gather information about enemy military units in the region.

• Request additional imagery and deploy an MQ-9 surveillance drone to capture video of a potential threat.

• Generate three courses of action (COAs) to target enemy equipment: COA 1 - send in an F-16 fighter jet, COA 2 - use long-range artillery, COA 3 - send in a tactical team with portable shoulder-fired Javelin missiles.  

• Send each COA to a commander for review before proceeding.

• Generate a battle plan to send in a tactical team (COA 3) with enough ammunition to engage the enemy and the appropriate equipment to jam the enemy's communications systems.

Other examples of the current (and expected) use of AI and autonomous weapons in warfare:

• AI Drones in Ukraine/Russia: Military experts have raised concerns about the use of semi- and fully autonomous drones in the Russia-Ukraine war. The reportedly modified consumer quadcopter drones can drop grenades on enemy fighters and could be a precursor to other lethal autonomous weapons systems (or LAWS) with the ability to self-navigate, coordinate attacks, and detect targets.

• Poseidon Torpedoes Produced in Russia: According to an unidentified defense source cited by Russian state news agency TASS, the country has produced the first set of nuclear-capable Poseidon super torpedoes. Poseidon has been described as a cross between a torpedo and a drone that can be launched from a nuclear submarine, has unlimited range, can operate at extreme depths, and can create radioactive ocean waves that can render coastal cities uninhabitable.

• US DOD Revives JADC2: The U.S. Department of Defense (DOD) is reviving efforts to implement Joint All-Domain Command and Control (JADC2), which the DOD says will enable the group to understand and act on “information across the battle-space quickly using automation, AI, predictive analytics, and machine learning.” The JADC2 has also been described as a means to connect all armed forces, including the Navy, Air Force, Marines, and Space Force, "into a single cloud-like network," allowing teams to seamlessly share data and " deploy the full force of military capabilities during current and future conflict."  

Why this Matters:

• Autonomous weapons systems (AWS) could potentially revolutionize warfare in the same way that gunpowder and nuclear weapons changed the global balance of power, and the way countries strategize battlefield engagements. At best, autonomous weapons could be faster and more accurate than current weapons, potentially limiting the casualties of war while helping militaries effectively outmaneuver their enemies.

• However, there are still unresolved concerns about the ethical use of these systems, whether they comply with international humanitarian law, and whether they can be used safely and reliably.

• Systems that would rely on large language models (LLMs), such as Palantir's AIP, are of particular concern because LLMs tend to fabricate information, provide solutions that don't work upon further inspection, and struggle (or have limited performance) with basic math and reasoning tasks.

Additional Links for “Update #1. How Generative AI (GenAI) Could Redefine the Battlefield”:

• Generative AI News Headlines

• Military News Headlines

• Large Language Model News Headlines

Update #2 - Research. AI Systems that Can Generate Fashion Videos, Clean Rooms, and Learn from Multiple Sources.

DreamPose: In AstroFeather Newsletter Issue No. 3, I updated you on the role of AI in fashion. Since then, there's been an AI Fashion Week (AIFW) (with judges from Vogue and luxury fashion brand Celine), and Forbes recently reported that small and medium-sized businesses are turning to startups like Botika and LalalandAI for custom AI-generated models to showcase their clothing. There are also some interesting research projects, such as DreamPose, that are specifically targeting the fashion industry.

DreamPose is a generative AI (GenAI) model that transforms fashion images into photorealistic, animated videos to create a more immersive experience with clothing. While static images are limited in their ability to show how a garment fits and moves on a person's body, DreamPose enables the creation of high-quality fashion videos from a single input image and a pose sequence.

With this novel approach, brands and retailers can better showcase their products and provide consumers with more engaging and informative content to help them make more informed purchasing decisions. Despite some recent controversy, I believe that generative AI platforms like Botika, Lalaland, and perhaps one day DreamPose, will become mainstays in the fashion industry.

[ Project Page | GitHub Code | Research Paper ]

TidyBot: Imagine if you had a Roomba that could do more than just clean your floors. For example, you could ask it to help you tidy up your room with instructions like "put the clothes in the laundry basket" or "put the soda cans in the recycling bin," and it would do the tasks as requested.

Recently, a team of researchers from Princeton, Stanford, Columbia University, and Google accomplished something similar by developing a personalized robotic assistant called TidyBot that can sort laundry, recycle, and organize items based on a user's instructions.

The study explores the feasibility of robotic household cleaning using an AI system that combines object detection, image classification, and LLMs to interpret natural language commands (i.e., user preferences) and allows the robot to successfully organize 85% of objects in real-world scenarios.

[ Project Page | GitHub Code | Video ]

ImageBind: Meta recently unveiled ImageBind, an open-source AI model that combines six types of data, including text, audio, visual (image and video), depth information (3D), temperature (heat/thermal), and movement, into a single multidimensional index or "embedding space".

By linking these six different forms of information, ImageBind can mimic the way humans interpret the world, bringing AI one step closer to learning directly from many different forms of information.

It's an impressive project, and some example uses include generating audio samples based on an input image or video, converting audio clips to images, and converting a text description to an image with audio.

[ Demo | GitHub Code | Research Paper ]

Additional Links for “Update #2 - Research. AI Systems that Can Generate Fashion Videos, Clean Rooms, and Learn from Multiple Sources”:

• AI Research and Development News Feed.

• AI Fashion News Headlines.

Update #3. Product Launches and Updates from Google, IBM, and Anthropic.

During the recent Google I/O event, we got a look at Google's latest collection of new AI platforms, services, and upgrades to existing products. You can watch the full 2-hour presentation [ here ] or the 10-minute recap [ here ]. Below, I've listed some highlights from the event, including new LLMs, as well as AI upgrades to Bard, Gmail, and Workspace apps like Google Docs.

PaLM 2: Google unveiled PaLM 2, its latest large language model (LLM), which will power the updated Bard chat tool and serve as the underlying model for most of its new AI features. PaLM 2 reportedly has improved capabilities for reasoning, mathematics, logic, and multilingual tasks, having been trained on a corpus of more than 100 languages.

During the Google I/O event, CEO Sundar Pichai stated that PaLM 2 will be available in different sizes called Gecko, Otter, Bison, and Unicorn. According to Pichai, the Gecko model is "so lightweight it can work on mobile devices [and is] fast enough for great interactive applications on device, even when offline."

The PaLM family of LLMs also includes Codey for code generation, Med-PaLM 2 fine-tuned for the medical field, and Sec-PaLM for security use cases.

Google Bard Gets Smarter: Google and Adobe have teamed up to integrate Firefly, Adobe's AI image generator, into Google Bard. With this new upgrade, Bard users will be able to create and edit images, as well as incorporate them into designs using Adobe Express.

According to Adobe, users do not need to worry about copyright issues because Firefly uses licensed images from Adobe Stock to train its model, making it free of copyrighted material such as branded content and resulting in images that are safe to use in a commercial environment.

The Bard AI chatbot will also include support for new languages such as Japanese and Korean, as well as the ability to export text to Google Docs and Gmail, perform visual searches, debug, and explain chunks of code, and integrate with third-party web services such as Instacart and OpenTable.

Duet AI Brings AI tools to Gmail, Google Docs and More: Google's Duet AI offers several generative AI tools such as writing assistance and proofreading in Google Docs and Gmail, automatic meeting summaries for Google Meet, image generation for Google Slides, and project organization in Google Sheets.

The company also recently announced the launch of "Help me write" - an upgrade to Smart Compose - as a writing assistant for Gmail on mobile. However, many of these features are still in development and only available through the Workspace Labs waiting list.

IBM watsonx: IBM has launched watsonx, a new AI and data platform designed to help businesses integrate AI into their operations. The technology is IBM's attempt to win new business after its earlier Watson software proved too expensive for many companies to deploy. IBM says watsonx can be used to train and deploy AI models and generate code using natural language.

Anthropic’s Claude Can “Remember” Entire Novels: Anthropic, an AI startup founded by former OpenAI engineers, has expanded the context window of its chatbot Claude to around 75,000 words (or 100,000 tokens). This is a significant improvement over current models, such as OpenAI's GPT-4, which processes around 8,000 to 32,000 tokens.

With this new capacity, Claude can process an entire novel like The Great Gatsby in less than a minute, digest and summarize dense documents like financial statements and research papers, as well as hold longer conversations.

Additional Links for “Product Launches and Updates from Google, IBM, and Anthropic”:

• Product Launch News Feed.

• Google Product and Services News Headlines.

Update #4. Company Announcements from Wendy’s, Anthropic, Google, and Amazon.

Wendy’s AI Drive-Thru Deal with Google: Wendy's is set to roll out an AI chatbot for its drive-thru service, with the help of Google. The technology will be rolled out at a single Wendy's restaurant in Columbus, Ohio, and will feature unique menu words, item names, and acronyms specific to the franchise.

To deliver quality results, the AI chatbot must also understand different dialects and accents, while cutting through background noise such as music or people chatting in the backseat.

Anthropic Proposes New AI Training Methods: Anthropic has developed a "constitutional AI" training approach that gives its AI chatbot, Claude, explicit values. This approach trains AI language models to provide answers to adversarial questions, while conditioning them with a set of behavioral principles.

Anthropic's initial list of principles includes several trust and safety "best practices," parts of Apple's terms of service, and the United Nations Declaration of Human Rights. The constitution is still being finalized, and Anthropic plans to continue iterating based on feedback. Unlike reinforcement learning from human feedback (RLHF), this approach does not rely on human labor.

Google is in the Hearing Aid Business: Google is working with Cochlear and other organizations to improve existing hearing aid technology and develop new solutions for people with hearing loss. Google plans to use AI and machine learning to personalize the experience for each person and environment.

By identifying, categorizing, and separating sound sources, AI can help hearing aids and cochlear implants reduce background noise and make speech clearer. With more than 1.5 billion people worldwide currently living with hearing loss, protecting hearing is critical.

Amazon Burnham: Amazon is reportedly working on upgrading its Astro home robot with a conversational interface known internally as "Burnham." According to internal documents obtained by Business Insider, Project Burnham aims to make the robot more perceptive and intelligent, allowing it to automatically respond to observations and Q&A-style dialogues and then take appropriate actions.

Burnham is expected to use large language models (LLMs) to help the Astro robot system perform complex tasks in response to natural language commands. The Burnham project is another example of how Amazon is integrating generative AI and LLMs into its products and services.

Additional Links for “Update #4. Company Announcements from Wendy’s, Anthropic, Google, and Amazon”:

• Product Launch and Announcements News Feed.

• Venture Capital and Startups News Feed

Thanks for reading this issue of the AstroFeather newsletter!

Be sure to check out the AstroFeather site for daily AI news updates and roundups. There, you'll be able to discover high-quality news articles from a curated list of publishers (ranging from well-known organizations like Ars Technica and The New York Times to authoritative blogs like Microsoft's AI Blog) and get recommendations for additional news articles, topics, and feeds you might enjoy.

See you in the next issue!

Adides Williams, Founder @ AstroFeather (astrofeather.com)