Best of Generative AI — December 2024

  1. 1
    Article
    Avatar of swirlaiSwirlAI·1y

    Building AI Agents from scratch - Part 1: Tool use

    Learn how to build AI agents from scratch, focusing on implementing tool usage capabilities without any orchestration frameworks. The guide covers creating Python functions as tools, constructing effective system prompts, and developing an Agent class to plan and execute actions using provided tools. The tutorial includes detailed code examples and explanations for wrapping functions as tools, formatting prompts, and executing tasks effectively.

  2. 2
    Article
    Avatar of hnHacker News·1y

    Building AI Products—Part I: Back-end Architecture

    In 2023, an AI-powered Chief of Staff tool for engineering leaders reached 10,000 users within a year. Insights gathered during its development led to the creation of Outropy, a developer platform to build AI products, focusing on sustainable and reliable AI systems. The journey involved navigating challenges with generative AI, understanding the role of agents versus microservices, and optimizing performance and scalability. The transition to using Temporal for stateful workflows and the evolution of AI product development is a highlight, offering valuable lessons in structuring AI applications.

  3. 3
    Article
    Avatar of hnHacker News·1y

    OpenAI whistleblower Suchir Balaji found dead in San Francisco apartment

    Former OpenAI researcher and whistleblower Suchir Balaji was found dead in his San Francisco apartment. Balaji had previously accused OpenAI of violating U.S. copyright law in developing ChatGPT. His information was crucial for ongoing lawsuits against the company. Despite the mysterious circumstances, police found no evidence of foul play. Balaji argued that OpenAI's practices were unsustainable and harmful to the internet ecosystem.

  4. 4
    Article
    Avatar of dailydoseofdsDaily Dose of Data Science | Avi Chawla | Substack·1y

    Intro to ReAct (Reasoning and Action) Agents

    ReAct agents, combining the reasoning power of LLMs with actionable steps, enable AI to both plan and interact with the real world. By using the open-source, low-code framework Dynamiq, users can seamlessly build and manage AI agents. Dynamiq simplifies complex LLM workflows and supports RAG applications through an intuitive API. The post includes a detailed guide to building a ReAct agent, from installation to execution.

  5. 5
    Article
    Avatar of infoworldInfoWorld·1y

    How to build better AI chatbots

    Campfire launched Cozy Friends, an AI chat product, achieving over 1.7 million messages exchanged in its first 30 days. Key lessons learned include treating system prompts dynamically like React apps, opting for deterministic outcomes early, utilizing model blending, employing scripted responses, and crafting engaging conversation starters. Metrics for judging AI output are essential for maintaining quality across various interactions.

  6. 6
    Article
    Avatar of heatherbcooperVisually AI·1y

    My AI Favorites This Year: Game-Changing Tools of 2024

    Key AI tools of 2024 have significantly streamlined workflows. ChatGPT enhances research and content creation with features like web search and memory updates. Claude’s Pro plan offers custom prompts for various topics and visual information generation. Magnific excels in upscaling images, while FLUX provides an open-source alternative for image generation. AI video creation is revolutionized by various models like Gen-3 and Pika 2.0. Additionally, tools like Canva, CapCut, and Descript facilitate diverse content creation tasks.

  7. 7
    Article
    Avatar of communityCommunity Picks·1y

    Build Tool Calling Agents with LangGraph and watsonx.ai Flows Engine

    Generative AI allows for creating applications by connecting tools, models, and APIs. This tutorial shows you how to set up and run an AI agent using watsonx.ai Flows Engine and LangGraph, with models powered by IBM's watsonx.ai platform. The guide covers installation, tool setup, and running a chat application. Learn to convert your own data sources into tools and build AI-powered apps with real-time data retrieval.

  8. 8
    Article
    Avatar of infoqInfoQ·1y

    PydanticAI: A New Python Framework for Streamlined Generative AI Development

    PydanticAI, a new Python framework inspired by FastAPI, facilitates the development of production-ready Generative AI applications. It offers a type-safe, model-agnostic approach and supports multiple AI models. Key features include a type-safe dependency injection system, Logfire integration for debugging, and a Pythonic design for ease of use. PydanticAI aims to replicate the success and ergonomics of FastAPI in the generative AI space, drawing positive feedback for its robust features and excellent documentation.

  9. 9
    Article
    Avatar of ds_centralData Science Central·1y

    LLM 2.0, RAG & Non-Standard Gen AI on GitHub

    Vincent Granville discusses LLM 2.0, introducing innovative approaches to generative AI and LLM advancements. He highlights the limitations of traditional LLMs and presents new methodologies, including multi-tokens, knowledge graph tokens, and NoGAN for data synthesis. The post also covers his repository's open-source code, emphasizing its usefulness for enterprise applications with examples like the NVIDIA case study and xLLM built from the Wolfram corpus.

  10. 10
    Article
    Avatar of communityCommunity Picks·1y

    pydantic/pydantic-ai: Agent Framework / shim to use Pydantic with LLMs

    PydanticAI is a Python agent framework designed to simplify building GenAI applications, similar to how FastAPI revolutionized web development. It supports multiple LLM models, offers type-safe control flows, and integrates with Logfire for debugging and monitoring. PydanticAI is currently in early beta, and feedback is welcome.

  11. 11
    Article
    Avatar of auth0Auth0·1y

    RAG and Access Control: Where Do You Start?

    Developers of GenAI-powered applications must ensure the information provided complies with access control policies. Fine-grained authorization (FGA) helps safeguard data by only allowing authorized users to access specific information. Okta FGA enables developers to enforce complex authorization decisions in applications, using an example of a RAG-based AI assistant to demonstrate its use. The tutorial guides through setting up Okta FGA and OpenAI accounts and integrating them into an application to control data access based on user privileges.

See all Generative AI archives