# Apr 18, 2024

## Goldene: A single atom layer of gold
- **Linköping University researchers have created "goldene," a material consisting of gold sheets only a single atom layer thick**, which exhibits new properties potentially useful for carbon dioxide conversion and hydrogen production.

## Hardest problem in computer science: centering things
- Despite the **simplicity of CSS commands** for centering elements, a gap persists between **knowledge and application** in real-world interfaces, leading to widespread misalignment across major platforms.

## Embeddings are a good starting point for the AI curious app developer
- **Vector embeddings** transform complex human knowledge into **numerical arrays**, simplifying the integration of advanced search and recommendation features into applications by individual developers.

## Show HN: Speeding up LLM inference 2x times (possibly)
- The **Effort Engine demo** showcases the capabilities of the Effort algorithm, providing a **quick preview** and in-depth details at [kolinko.github.io/effort/](https://kolinko.github.io/effort/).

## Google is combining its Android and hardware teams – and it's all about AI
- **Google's new "Platforms and Devices" team**, led by Rick Osterloh, aims to **accelerate AI integration** across all its products, including Android, Pixel, and more, by consolidating hardware and software efforts.

## Cyc: History's Forgotten AI Project
- **Cyc, an AI project initiated in 1984, aimed to encode common-sense knowledge for human-like reasoning**, amassing a knowledge base of **1.5 million concepts and 25 million rules** to support inferences from extensive reasoning chains.

## Humane AI – Pico Laser Projection – AI Twist on an Old Scam (2023)
- **Humane AI's approach** combines outdated laser projection technology with AI hype, attempting to innovate on a concept that has previously failed, such as the projector phone, but without offering substantial improvements or addressing the inherent limitations of laser projection.

## Stable Diffusion 3 API Now Available
- **Stable Diffusion 3 and its Turbo version** are now accessible via the [Stability AI Developer Platform API](https://platform.stability.ai/docs/api-reference#tag/Generate/paths/~1v2beta~1stable-image~1generate~1sd3/post), boasting advancements in text-to-image generation that rival or surpass competitors like DALL-E 3 and Midjourney v6.

## Solving the minimum cut problem for undirected graphs
- **Google Research's Di Wang** and team have developed a **deterministic nearly-linear time algorithm** for the **min-cut problem** in weighted graphs, marking a significant advancement in algorithmic graph theory. [Read the paper](https://epubs.siam.org/doi/10.1137/1.9781611977912.111)

## Feds appoint “AI doomer” to run US AI safety institute
- **Paul Christiano**, a former **OpenAI researcher** known for pioneering **reinforcement learning from human feedback (RLHF)**, has been appointed to lead the **US AI safety institute**, despite his controversial view that AI development has a **50 percent chance of ending in 'doom'**.

## Meta releases Llama 3
- **Meta releases Llama 3**, an advanced iteration of its language model, aiming to enhance AI's understanding and generation of human language.

## Implementing Natural Conversational Agents with Elixir
- Sean Moriarity implemented **Nero**, an **Elixir-based conversational agent**, enhancing it for natural interaction by integrating **voice activity detection (VAD)**, **speech-to-text**, and **text-to-speech** technologies, notably using **Nx, OpenAI APIs, and ElevenLabs**.

## How Antithesis finds bugs
- **Antithesis utilizes a deterministic hypervisor** to explore complex system state spaces, efficiently finding bugs by leveraging strategies from playing Nintendo games, notably **Super Mario Bros.**

## USAF Test Pilot School, DARPA announce aerospace machine learning breakthrough
- The **USAF Test Pilot School and DARPA** were finalists for the **2023 Robert J. Collier Trophy**, recognizing their **breakthroughs in aerospace machine learning** with the **X-62A VISTA aircraft** under the **Air Combat Evolution (ACE) program**.

## What comes first, math, or algorithm in research?
- **Research in machine learning** often oscillates between **algorithmic innovation** and **mathematical foundation**, with some researchers starting from a practical problem and then seeking mathematical explanations, while others apply existing mathematical theories to generate new algorithms.