In partnership with

Hello readers, in this issue we cover

• Stable Diffusion 3.5 is released. These models focus on customizability and performance efficiency

• Google can watermark AI generated text

• U.S. spurs increased investment in AI in defense

• Meta gives robot touch with multimodal, artificial fingerprints

• AI creates more efficient medical workflows by aiding humans in clinical coding

Learn AI in 5 minutes a day.

The Rundown is the world’s most trusted AI newsletter, with over 700,000+ readers staying up-to-date with the latest AI news, understanding why it matters, and learning how to apply it in their work.

Their expert research team spends all day learning what’s new in AI, then distills the most important developments into one free email every morning.

Sign up.

📸 Stable Diffusion 3.5 has arrived

StabilityAI announced the release of their latest, most powerful set of models yet:

• Stable Diffusion 3.5 Large with 8.1 billion parameters and produces images of superior quality and prompt adherence

• Stable Diffusion 3.5 Large Turbo is a distilled, faster version of 3.5 Large

• Stable Diffusion 3.5 Medium with 2.5 billion parameters designed to be run out-the-box on consumer hardware

When developing the models, StabilityAI prioritized customizability and efficient performance.

Read the announcement

💧 Google watermarks AI-generated text

LLM can create realistic, human-like text. To prevent misuse, watermarking can help identify synthetic text, but existing methods haven’t met the quality and efficiency needed for real-world use. SynthID-Text is a new watermarking system that maintains text quality and accurately detects watermarked content without slowing down the LLM. It works by adjusting the text generation process rather than changing the model itself, making it efficient and production-ready. Tested on various models, SynthID-Text showed high detection accuracy and no drop in performance. A large-scale test with millions of responses confirmed its effectiveness, aiming to encourage responsible LLM use.

Read the research paper

🇺🇸 US to increase investment in AI research for national security

On October 24, 2024, The White House issued a memorandum to enhance U.S. leadership in artificial intelligence (AI), focusing on national security and ethical considerations. The directive emphasizes the rapid and responsible deployment of AI technologies to maintain a competitive edge, particularly against nations like China. It mandates federal agencies to strengthen AI-related supply chains, ensure AI systems align with democratic values, and implement safeguards against misuse. The memorandum also calls for collaboration with international allies to establish global AI standards and underscores the importance of human oversight in AI applications.

Read the research paper

👆Meta gives robots artificial multimodal fingerprints

Touch is an important sense for both humans and robots, providing valuable information about objects and interactions with the world. This new research presents a highly advanced artificial fingertip that can sense touch with great detail, similar to human touch but with even more precision. This fingertip includes millions of tiny sensors that detect small details, different forces, heat, vibration, and even odor. It also has a built-in AI system that processes this information instantly, like a human’s reflexes. The technology could be revolutionary for robots, VR, prosthetics, and more, and the researchers are making their platform available for others to build on.

Read the research paper

🩻 AI improves clinical coding to ICD-10 codes

This study tested an AI tool, Easy-ICD, to help clinical coders improve accuracy and speed. In a trial with participants in Norway and Sweden, coders were divided into two groups and switched between coding complex (long) and simple (short) texts, using the AI tool for some tasks but not others. Results showed that using Easy-ICD reduced the time needed to code complex texts by 46%, but it didn't significantly affect the time for simpler texts or accuracy for either type. The study suggests that AI could improve efficiency in complex coding tasks, though more research is needed in real hospital settings.

Read the research paper

🤯 Today I Learned

Every issue, we highlight new AI concepts and terminology to help educate our readers. This issue we learned about:

Diffusion Model

A diffusion model is a type of probabilistic machine learning model that learns to generate data by simulating a gradual noise-removal process. It’s commonly used for creating images, audio, and other types of data by reversing a process that adds noise in small steps until the original data becomes pure noise. By learning to reverse this process step-by-step, the model can start with random noise and iteratively “denoise” it to produce realistic, high-quality samples.

ICD-10 code

An ICD-10 code is a code from the International Classification of Diseases, 10th Revision, which is a medical classification system developed by the World Health Organization (WHO). Each code represents a specific diagnosis, condition, or symptom and is used worldwide for standardized reporting and billing purposes in healthcare. ICD-10 codes allow healthcare providers, insurers, and researchers to communicate and analyze health information consistently.

These codes are crucial for medical records, insurance claims, and public health tracking, as they ensure consistent and accurate descriptions of medical conditions across different countries and healthcare systems.