# Aug 14, 2024

### AudioFlux: A C/C++ library for audio and music analysis
- **audioFlux** is a **deep learning library** designed for **audio and music analysis**, offering extensive **feature extraction** capabilities across various time-frequency analysis methods.

### Grok-2 Beta Release
- **Grok-2 outperforms Claude 3.5 Sonnet and GPT-4-Turbo** on the LMSYS leaderboard, showcasing advancements in chat, coding, and reasoning capabilities.

### MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Images
- **MVSplat** introduces an **efficient model** that predicts clean feed-forward **3D Gaussians** from sparse multi-view images, leveraging a **cost volume representation** for accurate localization.

### ARPA-H announces awards to develop novel technologies for precise tumor removal
- **ARPA-H's PSI program** awards **$150 million** to develop technologies for **precise tumor removal**, aiming to **reduce repeat surgeries** and **unintentional injuries** during cancer operations. [PSI program page](https://arpa-h.gov/research-and-funding/programs/psi)

### [R] The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery
- **The AI Scientist** framework enables **frontier large language models** to independently conduct scientific research, **generating novel ideas**, writing code, executing experiments, and **communicating findings** through full scientific papers, including a simulated review process. [Paper](https://arxiv.org/abs/2408.06292)

### "Mutual Reasoning" improves GSM8K accuracy from 13% to 64% [R]
- **rStar**, a **self-play mutual reasoning approach**, significantly **enhances the reasoning capabilities** of small language models (SLMs) by employing a novel generation-discrimination process without the need for fine-tuning or larger models.

### Wet-lab innovations will lead the AI revolution in biology
- **Abhishaike Mahajan** specializes in **biology posting** on Substack, offering insights into biology and machine learning through essays.

### [R] Grokfast: Accelerated Grokking by Amplifying Slow Gradients
- **Grokfast** significantly **accelerates the grokking phenomenon** in machine learning by focusing on **amplifying slow-varying gradient components**, which are crucial for generalization. [Read the paper](https://arxiv.org/abs/2405.20233v2)

### [R] Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2
- **Gemma Scope introduces an open suite of JumpReLU Sparse Autoencoders (SAEs)** trained across various layers of Gemma 2 models, aiming to democratize access to advanced neural network interpretability tools.

### [R] New Paper on Mixture of Experts (MoE) 🚀
- The new paper on **Mixture of Experts (MoE)** delves into **advancements** in balancing **computational efficiency** with **high performance** in AI systems, addressing **current challenges** and **future directions**.

### [P] New open-source release: SOTA multimodal embedding models for fashion
- **Marqo-FashionCLIP & Marqo-FashionSigLIP**, new **open-source multimodal models**, have **outperformed existing SOTA models** like FashionCLIP2.0 and OpenFashionCLIP on **7 fashion evaluation datasets** including DeepFashion and Fashion200K by **up to 57%**.

### LLMs as Optimizers - Theory Paper Recommendation [R]
- **Transformers** are theorized to perform **first-order optimization**, such as gradient descent, expanding their utility beyond traditional applications.

### [R] Ensemble everything everywhere: Multi-scale aggregation for adversarial robustness
- The proposed **multi-resolution input representations and dynamic self-ensembling** technique enhances **adversarial robustness** by leveraging intermediate layer predictions, which are naturally resistant to attacks.

### [D] opinions on keras 3 with jax
- The **interest in JAX** and its ecosystem, including **Flax (linen and nnx)**, stems from its **mathematical simplicity** and **functional programming** approach, contrasting with TensorFlow's complexity.

### Decoding NVIDIA Edify — The Technology That Helps Developers Create Custom Models Trained on Their Data
- **NVIDIA Edify** enables developers to create **custom generative AI models** using their own data, supporting a wide range of content including images, videos, and 3D assets.