Apr 23, 2024

• Ex-Amazon exec claims she was asked to ignore copyright law in race to AI
  Dr. Viviane Ghaderi filed a lawsuit against Amazon, alleging she was demoted and dismissed after her pregnancy leave, claiming discrimination and urging her to ignore copyright laws in AI research.

• Phi-3 Technical Report
  Phi-3-mini, a 3.8 billion parameter language model, achieves competitive performance on benchmarks like MMLU (69%) and MT-bench (8.38), rivaling larger models while being efficient enough for phone deployment.

• Dify, a visual workflow to build/test LLM applications
  Dify is an open-source LLM app development platform that integrates AI workflow, RAG pipeline, agent capabilities, and model management, facilitating the transition from prototype to production.

• Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
  Phi-3-mini, a 3.8 billion parameter language model, achieves competitive performance with larger models like GPT-3.5, scoring 69% on MMLU and 8.38 on MT-bench, and is optimized to run on mobile devices. Read more

• Why isn't GNN in high demand in industry?
  Graph Neural Networks (GNNs) are not widely sought after in industry roles due to their computational demands and the complexity of preprocessing data into graph formats.

• Parquet-WASM: Rust-based WebAssembly bindings to read and write Parquet data
  parquet-wasm offers WebAssembly bindings for converting between Apache Parquet and Apache Arrow formats, leveraging Rust's parquet and arrow crates, designed to work with JavaScript Arrow implementations.

• Espressif Acquires Majority Stake in M5Stack
  Espressif Systems acquires a majority stake in M5Stack, aiming to enhance the AIoT ecosystem by integrating M5Stack's modular, open-source platform with Espressif's technology.

• FPGA Architecture for Deep Learning: Survey and Future Directions
  FPGAs offer a unique blend of reprogrammability and hardware execution, making them ideal for accelerating deep learning (DL) inference with the ability to customize processing pipelines for lower latency and higher energy efficiency.

• When will computer hardware match the human brain? (1998)
  Hans Moravec predicts that by the 2020s, affordable machines will possess the hardware capacity to match human intellectual performance, based on the extrapolation of past trends in computing power and the development of AI technologies.

• Modal is a matrioshka language based on pattern-matching to rewrite trees
  Modal is a matrioshka language designed for rewriting trees through pattern-matching, utilizing substitution rules that transform a given tree until no rules apply.

• How to Deploy LLaMA 3 Into Production, and Hardware Requirements
  Deploying LLaMA 3 on an AWS EC2 instance involves distinct requirements for the 8B and 70B versions, with a comprehensive tutorial available at nlpcloud.

• Generative A.I. arrives in the gene editing world of CRISPR
  Generative A.I. technology, developed by Profluent, creates blueprints for gene editors like OpenCRISPR-1, capable of editing human DNA, marking a significant leap in the precision and speed of battling illnesses.

• Survey Study on AI Agent Architectures (2024)
  This survey paper highlights recent advancements in AI agent implementations, focusing on enhanced reasoning, planning, and tool execution capabilities to achieve complex goals.

• LLMs and the Harry Potter problem
  Large language models (LLMs) struggle with tasks requiring in-context recall and counting in large documents, as demonstrated by their performance on a test involving counting the word "wizard" in a chapter of Harry Potter, with accuracy rates significantly varying across models.

• It's Hyperobjects All the Way Down
  Timothy Morton introduces the concept of hyperobjects to describe phenomena too vast for human comprehension, such as climate change and pandemics, challenging our conventional understanding of reality.