# Nov 18, 2024ML Times

## - All-in-one embedding model for interleaved text, images, and screenshots
  
- **voyage-multimodal-3** is a groundbreaking multimodal embedding model that vectorizes interleaved text and images, enhancing retrieval accuracy by an average of **19.63%** over its closest competitor across various tasks, including document screenshots and table/figure retrieval.

## - Qwen2.5 Turbo extends context length to 1M tokens

- **Qwen2.5-Turbo** now supports a **context length of 1M tokens**, enabling the processing of vast amounts of text, equivalent to **10 full-length novels** or **150 hours of speech**, while achieving **100% accuracy** in specific tasks and outperforming competitors like GPT-4 in benchmarks.

## - Garak, LLM Vulnerability Scanner

- **`garak` is a comprehensive LLM vulnerability scanner** that identifies weaknesses such as hallucination, data leakage, and prompt injection, functioning similarly to `nmap` for network security but tailored for language models.

## - You could have designed state of the art positional encoding

- The post details the **evolution of positional encoding** in transformer models, culminating in **Ro** tary **P** ostional **E** ncoding (RoPE), which enhances self-attention mechanisms by encoding relative positions through multiplicative rotations rather than additive methods.

## - Two Nobel Prize winners want to cancel their own CRISPR patents in Europe

- **Nobel laureates Emmanuelle Charpentier and Jennifer Doudna are seeking to cancel their own CRISPR patents in Europe** to avoid the repercussions of a recent unfavorable ruling that questioned the validity of their invention, which could reshape the landscape of CRISPR licensing and ownership.

## - GaussianAnything: Interactive Point Cloud Latent Diffusion for 3D Generation

- **GaussianAnything** employs a **cascaded 3D diffusion pipeline** to produce _high-quality_ and _editable_ surfel Gaussians from single-view images or text inputs, enhancing the versatility of 3D generation techniques.

## - LLaVA-O1: Let Vision Language Models Reason Step-by-Step

- **LLaVA-o1** introduces a **novel approach** to Vision-Language Models (VLMs) by enabling **autonomous multistage reasoning**, significantly enhancing performance in complex visual question-answering tasks.

## - AMD Now Has More Compute on the Top500 Than Nvidia

- **AMD has surpassed Nvidia** in compute power on the Top500 list, primarily due to the performance of the "El Capitan" supercomputer, which achieved a peak theoretical performance of **2,746.4 petaflops** with its AMD MI300A devices.

## - Why LLMs Within Software Development May Be a Dead End

- **LLMs may hinder software development** due to their limitations in understanding complex coding contexts and producing reliable outputs, which can lead to increased debugging time and reduced productivity.

## - Electron Spins: a special case of Chromium mods

- **Electron spins** in Chromium mods present unique properties that enhance quantum computing applications, particularly in **spintronic devices** which leverage electron spin for data processing.

## - [R] Performance Analysis of GPU Interconnect Technologies Across Three Modern Supercomputer Architectures

- **NVLink** offers the **highest bandwidth** (up to **900 GB/s**) for GPU communication, but incurs **higher latency**, making it essential to evaluate workload needs when selecting interconnects.

## - A Taxonomy of AgentOps

- **Foundation model (FM)-based autonomous agents** are increasingly complex, necessitating a shift towards **AgentOps platforms** that prioritize **observability** and **traceability** throughout their life-cycle.

## - AI Will Drive Scientific Breakthroughs, NVIDIA CEO Says at SC24

- **NVIDIA's innovations in AI and supercomputing are set to revolutionize industries**, with tools for drug discovery, climate forecasting, and quantum simulations, showcasing a commitment to driving scientific breakthroughs.

## - Faster Forecasts: NVIDIA Launches Earth-2 NIM Microservices for 500x Speedup in Delivering Higher-Resolution Simulations

- **NVIDIA's new NIM microservices** enable a **500x speedup** in delivering high-resolution weather simulations, enhancing AI model deployment for predicting extreme weather events like snow, ice, and hail.

## - NVIDIA Releases cuPyNumeric, Enabling Scientists to Harness GPU Acceleration at Cluster Scale

- **cuPyNumeric** allows researchers to run their existing Python code on **thousands of GPUs** without any modifications, significantly enhancing data processing speed and efficiency across various scientific fields.