# Jul 8, 2024  
### Daily

#### - Reasoning in Large Language Models: A Geometric Perspective  
  - The **expressive power** of large language models ( **LLMs**) is intricately linked to the **density of their self-attention graphs**, which determines the **intrinsic dimension** of inputs, enhancing their reasoning capabilities.

#### - LivePortrait: A fast, controllable portrait animation model  
  - **LivePortrait introduces an efficient method for animating portraits**, leveraging advanced stitching and retargeting techniques to control animations dynamically.

#### - Show HN: A fast OSS voice assistant  
  - **Swift** is a **fast, open-source voice assistant** that leverages the power of [Groq](https://groq.com/), [Cartesia](https://cartesia.ai/), [VAD](https://www.vad.ricky0123.com/), and [Vercel](https://vercel.com/) for enhanced speech detection capabilities.

#### - [R] Learning to (Learn at Test Time): RNNs with Expressive Hidden States  
  - The **Test-Time Training (TTT) layers** introduce a novel approach by making the hidden state a **machine learning model itself**, enhancing expressiveness and performance in long contexts.

#### - Micro-agent: make an AI write code until it passes an unit test  
  - **Micro Agent** is an AI tool designed to write and iteratively fix code until all test cases pass, streamlining the development process by automating code generation and correction.

#### - [R] An Empirical Study of Mamba-based Language Models (8B Mamba-2-Hybrid on 3.5T tokens data)  
  - **Mamba-based models**, particularly the **8B Mamba-2-Hybrid**, outperform 8B Transformer models across **12 standard tasks** by an average of **+2.65 points**, showcasing superior efficiency and effectiveness in language modeling. [Study details](http://arxiv.org/abs/2406.07887)

#### - Learning to (Learn at Test Time): RNNs with Expressive Hidden States  
  - **Test-Time Training (TTT) layers** introduce a novel approach by making the hidden state a **machine learning model itself**, enhancing expressiveness and performance in long contexts.

#### - [R] A Universal way to Jailbreak LLMs' safety inputs and outputs if provided a Finetuning API  
  - A researcher has developed a **universal method to bypass safety checks** in Large Language Models (LLMs) using a **finetuning API**, employing a Caesar Cipher with 25 shifts to encode harmful instructions that the LLMs' safety mechanisms fail to detect.

#### - [R] What is GraphRAG? Explained  
  - **GraphRAG** represents an **advancement over the baseline RAG** by utilizing **Knowledge Graphs** for retrieval, which enhances the **quality of its output**.

#### - [P] ReproModel: Open Source ML Research Toolbox Update!  
  - **ReproModel**, an **open-source toolbox**, aims to **simplify the testing and reproduction of machine learning models**, addressing common issues like missing code and unclear experiment parameters.

#### - [Research] Neural decoding - mapping EEG data of song listening to respective audio files  
  - The **research aims to reconstruct songs** participants listened to by mapping **EEG data to audio files** using a regression model, with a **CNN-based approach** detailed in a [study](https://arxiv.org/abs/2207.13845).

#### - TwoMinutePapers - DeepMind’s New AI Found The Sound Of Pixels!  
  - **DeepMind's new AI technique** synthesizes sound by analyzing video content, a significant leap towards more immersive AI-generated media.

#### - In It for the Long Haul: Waabi Pioneers Generative AI to Unleash Fully Driverless Autonomous Trucking  
  - **Waabi** is leveraging **generative AI** and **NVIDIA's DRIVE Thor technology** to revolutionize autonomous trucking, aiming for **fully driverless operations** by next year.

#### - YannicKilcher - Scalable MatMul-free Language Modeling (Paper Explained)  
  - Researchers from **UC Santa Cruz, Sucha University, UC Davis, and Loxy Tech** have developed a **scalable, MatMul-free language model** that replaces matrix multiplication in large language models with ternary accumulators and a parallelizable form of a ternary recurrent network, aiming for **greater hardware efficiency**.

#### - ANAH-v2: Scaling Analytical Hallucination Annotation of Large Language Models  
  - The paper introduces **ANAH-v2**, an **iterative self-training framework** designed to scale and improve the accuracy of **hallucination annotation** in large language models (LLMs) by leveraging the Expectation Maximization algorithm.