Generative AI and LLM Insights: May 2024

The AI landscape is exploding in size, with some early winners emerging, but RAG reigns supreme for enterprise LLM systems. Check out our roundup of the top generative AI and LLM articles for May 2024.

The 2024 MAD (Machine Learning, AI & Data) Landscape

FirstMark Capital has identified 2,000+ companies (!!!) competing in the ML, AI, and Data world. Read an in-depth analysis to understand all the nuances in this complex space: https://mattturck.com/mad2024/

Current MAD landscape

A Survey on Retrieval-Augmented Text Generation for Large Language Models

RAG underpins many enterprise LLM systems. This recent paper consolidates existing research on RAG, explores its evolution, and suggests new areas for innovation and development: https://arxiv.org/abs/2404.10981

Basic RAG Workflow

7 Golden Rules for Generative AI Apps: A Playbook from Early Winners

AI-native enterprise apps have begun to find success, with the winners scaling even quicker than top SaaS companies. Menlo Ventures identified 7 common traits among these winners, providing a guide for all GenAI companies: https://menlovc.com/perspective/7-golden-rules-for-generative-ai-apps-a-playbook-from-early-winners/

In venture, software companies are often benchmarked against a best-in-class growth trajectory described as “triple, triple, double, double, double” after hitting $1 million ARR. But in the new era of generative AI, this benchmark falls short, only describing the lowest-quartile performance of breakouts from the generative AI wave.

How AI is helping the National Guard transform disaster relief

Climate change has made extreme weather events more common. Google's moonshot factory X is using AI to analyze disaster scenes in seconds, identify critical infrastructure, and create labeled maps to improve disaster response: https://x.company/case-study/bellwether-diu/

Analyzing a disaster scene using AI

The Unreasonable Ineffectiveness of the Deeper Layers

Unlike double stuff oreos, LLMs can have too many layers! Using unique pruning techniques, researchers removed half of the layers in an LLM with minimal impact on performance: https://arxiv.org/abs/2403.17887

Overview of layer-pruning strategy and example results

Advanced Retriever Techniques to Improve Your RAG

The "Retrieval" part of RAG is often overlooked. Learn some advanced retriever techniques to help you build a highly effective RAG system: https://towardsdatascience.com/advanced-retriever-techniques-to-improve-your-rags-1fac2b86dd61

Advanced retriever techniques