# Jun 29, 2024

- Microsoft says that it's okay to steal web content it because it's 'freeware.'

## News Highlights

- **Microsoft's AI CEO, Mustafa Suleyman, claims content on the web is 'freeware'** for AI training unless explicitly stated otherwise, sparking debate on the ethics and legality of using web content to train AI models.

- **Open-LLM performances have hit a plateau**, indicating a slowdown in the pace of advancements in large language models as observed by the Hugging Face Space's latest analysis.

- **EvTexture enhances video super-resolution** by integrating event-driven texture information, achieving significant improvements on the Vid4 and REDS4 test sets with a $4\times$ upsampling rate. [Paper](https://arxiv.org/abs/2406.13457)

- **GEMM tuning on AMD MI300x GPUs** significantly enhances throughput and latency for AI models, with throughput improvements up to **7.2x** for models like LLaMA-2-70B.

- **`llama-agents` is an async-first framework** designed for building and deploying multi-agent systems, facilitating features like multi-agent communication and human-in-the-loop processes.

- **Microsoft AI's CEO, Mustafa Suleyman, declared online content as 'freeware' for training AI models**, sparking legal challenges from content creators and organizations like the Center for Investigative Reporting and the New York Times.

- **Finetuning Large Language Models (LLMs) on a synthetic dataset** significantly enhances their ability to retrieve information and reason over long-context inputs, as demonstrated in experiments with GPT-3.5 Turbo and Mistral 7B.

- Microsoft has unveiled the **Skeleton Key attack**, a method that can bypass AI model safety measures to generate harmful content, such as instructions for making a Molotov cocktail.

- **Context-augmented Retrieval (CAR)** enhances Large Language Model (LLM) performance by partitioning the vector database based on real-time classification, ensuring quick and relevant information retrieval.

- The **Segment any Text (SaT) model** introduces a new pretraining scheme to enhance robustness by reducing reliance on punctuation, addressing a common shortfall in existing sentence segmentation methods.

- The **Vision Language Group at IIT Roorkee** has crafted detailed summaries of deep learning papers from 2016 to 2024, covering major conferences like NeurIPS, CVPR, ICCV, and ICML.