# May 16, 2024

- **Apple introduces new accessibility features** including Eye Tracking for iPad and iPhone, Music Haptics for the deaf or hard of hearing, and Vocal Shortcuts, leveraging **Apple silicon, AI, and machine learning** to enhance device usability for everyone.

- **Llama 3 model**, implemented in pure NumPy, aims to demystify its structure and operation, leveraging the **[stories15M model](https://github.com/karpathy/llama2.c?tab=readme-ov-file#models)** trained by Andrej Karpathy for an intuitive understanding.

- **Datomic Pro 1.0.7075** exhibits **inter-transaction Serializability** and **intra-transaction concurrent semantics**, ensuring operations within a transaction appear to execute simultaneously.

- **The pull request by jart introduces vectorized expf() functions to compute softmax and silu more accurately**, replacing the short[65536] lookup table previously used in GGML for speed, supporting aarch64 and sse2+ with a worst-case rounding error of 2 ulp.

- **Project Gameface**, an open-source, hands-free gaming 'mouse' that allows control via head movement and facial gestures, **launches on Android**, enhancing accessibility for people with disabilities.

- CPUs have **multiple cache levels** to balance speed, size, and power efficiency, with **L1 caches being smallest and fastest**, and higher levels increasing in size but decreasing in speed.

- **`llama3.np`** is a **NumPy-based implementation** of the Llama 3 model, demonstrating its functionality by running the stories15M model developed by Andrej Karpathy.

- The **EU has initiated formal investigations into Facebook and Instagram** over concerns that their design may be addictive to children, following a series of information requests to Meta under the Digital Services Act (DSA).

- **ChatGPT-4o's performance in solving a math problem varies significantly** across different experimental setups, including prompt-only, image inclusion, and the use of prompt engineering techniques like Chain-of-Thought.

- **Desktop Docs** simplifies file management by allowing users to **browse, edit, and export media files** using AI to search content within images and videos.

- Google's AI search feature, **uninvited and cumbersome**, introduces a **three-second delay** in search results, frustrating users with irrelevant AI-generated content.

- **Tarsier** is a **vision utility** designed to enhance web interaction agents by visually tagging interactable elements, enabling **LLMs to perform actions** on web pages more effectively.

- The **lack of implementation codes** in face anti-spoofing research papers **hinders reproducibility**, a cornerstone in the field of machine learning.

- **Mozilla Firefox** has integrated **NVIDIA RTX Video**, leveraging AI to upscale video quality and add HDR enhancements for users with RTX GPUs, enhancing streaming experiences on platforms like YouTube and Disney+.

- **Elicit automates research tasks** such as summarizing academic papers, extracting data, and synthesizing findings, catering to a database of **125 million papers**.