# Aug 6, 2024

### A New Type of Neural Network is More Interpretable

- **Kolmogorov-Arnold Neural Networks** (KANs) offer a **more interpretable framework** for AI, potentially guiding physicists towards novel hypotheses.

### WD Announces Enterprise 128TB SSD

- **WD announced a 128TB enterprise SSD, 8TB SD cards, and a 16TB external SSD** at FMS 2024, showcasing advancements in storage technology across various sectors.

### A RoCE Network for Distributed AI Training at Scale

- Meta has developed a **RoCE network infrastructure** to interconnect tens of thousands of GPUs for **large-scale distributed AI training**, supporting models with hundreds of billions of parameters like [LLAMA 3.1 405B](https://ai.meta.com/blog/meta-llama-3-1/).

### AI Agents but They're Working in Big Tech

- **AI multi-agent systems modeled after big tech companies' organizational structures, like Microsoft and Apple, show improved performance in software engineering tasks**, suggesting competitive team structures enhance problem-solving capabilities.

### Introducing the Open Medical Reasoning Tasks Project

- The **Open Medical Reasoning Tasks** project, initiated by Open Life-Science AI and inspired by NousResearch, aims to **develop benchmarks and datasets** for medical AI, focusing on the complex reasoning required in healthcare.

### Grounded SAM 2: Ground and Track Anything

- **Grounded SAM 2** enhances **video object segmentation and tracking** by integrating the **Grounding DINO** open-set detection model, expanding beyond its predecessor's capabilities.

### I Built an Open-Source Tool That Lets You Build GPU-Accelerated NNs and Transformers Directly on the Web

- **[JS-PyTorch](https://github.com/eduardoleao052/js-pytorch)** allows for **PyTorch-like code execution in web browsers**, leveraging JavaScript for an intuitive experience akin to the beloved PyTorch syntax.

### Diffusion vs Flow

- **Frontier labs** are increasingly adopting **flow-based methods** for **image and audio generation**, moving away from diffusion techniques.

### RAG Foundry: A Framework for Enhancing LLMs for Retrieval Augmented Generation

- **RAG Foundry** is an **open-source framework** designed to simplify the creation and evaluation of **Retrieval-Augmented Generation (RAG)** systems by integrating data creation, training, inference, and evaluation into a unified workflow.

### State of the Art in Scene/3D Model Generation from 2D Images

- **Scene reconstruction and 3D model generation from 2D images** are advancing rapidly, with **numerous papers and architectures** focusing on improving accuracy and efficiency.

### InternVideo2: An Open-source Video Understanding Model

- **InternVideo2** is an **open-source AI model** designed for **video understanding**, featuring a **6B parameter encoder** and leveraging over **400M+ samples** for superior dynamic scene perception and temporal reasoning.

### Exploring SELF-ROUTE: A Hybrid Approach to Efficient Long Context Question-Answering

- **SELF-ROUTE** combines **Retrieval-Augmented Generation (RAG)** and **Long Context (LC) question-answering**, leveraging **LLM self-reflection** to efficiently direct queries, achieving **cost savings** while preserving LC-like performance.

### Operationalizing Contextual Integrity in Privacy-Conscious Assistants

- **Operationalizing contextual integrity (CI)** in AI assistants involves designing strategies to ensure their information-sharing actions align with user privacy expectations, addressing the challenge of assistants accessing sensitive user data.

### TwoMinutePapers - OpenAI’s DALL-E 3-Like AI For Free, Forever!

- **Flux**, a new **text-to-image AI system**, rivals the capabilities of **DALL-E 3 and Midjourney**, offering **photorealistic images** and improved text generation within images, setting a new benchmark in AI-driven creativity.

### YannicKilcher - Privacy Backdoors: Stealing Data with Corrupted Pretrained Models

- Researchers from ETH Zurich have demonstrated a method for **extracting fine-tuning data from machine learning models** by corrupting pre-trained models, notably affecting models like BERT and ViTs, which are widely used in current applications.