GenerativeAI

How It Works

Applications, Stacks, Tools & Agents

Full Library

The models underpinning GenAI—like GPT-4 (OpenAI), Gemini (Google), and Llama 2 (Meta)—are often referred to as “foundation models.” The term "foundation model" was defined in 2021 as "any model that is trained on broad data (generally using self-supervision at scale) that can be adapted (e.g., fine-tuned) to a wide range of downstream tasks."*

‍

Adaption often takes the form of using models to power applications. Applications integrate the models with complementary tech and data in order to make them fit to purpose. The original ChatGPT, for example, was an application of a model, GPT 3.5, integrated with a complementary chat interface.

‍

Much of what is occurring in the world of GenAI involves the rapid maturing of a nascent application layer. Which is why over-indexing on the capabilities and limitations of the models in their raw form can lead the conversation astray.

‍

It is useful to understand that all GenAI models are part of a stack.* The models are run atop computing hardware, often in the cloud. The models are often connected to data sources through an orchestration layer that underpins an application layer—i.e., the layer at which end users interact not only with the model but also with the the other component parts. Further, a human services layer can be laid on top of the application layer.

‍

Part of building AI-enabled applications can include providing the AI with tools that extends its capabilities. For example, Advanced Data Analytics is a tool developed by OpenAI to complement GPT-4 and add data analytics capabilities that the model itself does not possess. A more digestible example is the likes of OpenAI, Microsoft, and Google giving their models access to web search (another tool).

‍

Providing GenAI applications access to tools is one step closer to the development of AI agents. The goal of AI agents is to be able to tell the AI what to do (“analyze this data”) rather than how to do it (“use Advanced Data Analytics to analyze this data”). That is, the human sets the objective, and the AI determines the methods as part of satisfying the objective. Quite cool. And also very scary.*

LLMs for Dummies

This article discusses the workings of Large Language Models (LLMs) in a layman's terms. LLMs represent words with numbers called word vectors, which are stored in clusters based on their relationships. The article explains how LLMs use transformer models to understand context and meaning, and how they improve with training. It also explores potential applications of LLMs, such as solving impossible problems, making easy problems convenient, vertical AI applications, and enhancing human productivity. The article concludes by highlighting the exponential growth and potential of LLMs.

Hide Summary

Ben Evans

Unbundling AI — Benedict Evans

ChatGPT 3.5 is a powerful technology that can answer questions and generate text. It has the potential to revolutionize product development. However, there are challenges with accuracy and presenting uncertainty. The use of LLMs raises science, product, and interface questions. It may be necessary to unbundle specific tasks into specialized tools. The future of LLMs is still uncertain, but they have the potential to transform various industries.

Hide Summary

North Western

October 12, 2023

The Half-Life of the AI Stack

BTSbot, an AI tool, automates supernova detection and classification, eliminating human intervention. Developed by an international team led by Northwestern University, it accelerates the analysis of supernova candidates.

Hide Summary

October 30, 2023

The architecture of today's LLM applications

Hide Summary

The Batch

September 20, 2023

Which AI Applications Should You Build? Here's How to Decide

AI applications should be decided by considering the tasks of the company's employees, assessing if they can be augmented or automated using AI tools, and evaluating the value of doing so. This approach focuses on automating tasks rather than entire jobs and can lead to revolutionary changes in businesses. AI will augment human labor and create new jobs, but society must support those affected by automation. Workers who know how to effectively use AI will replace those who don't.

Hide Summary

Honeycomb

May 26, 2023

All the Hard Stuff Nobody Talks About when Building Products with LLMs

In this post, Phillip talks through the challenges & pitfalls of LLMs we faced when building our Query Assistant - and that you too may face.

Hide Summary

Generative Conversations

August 25, 2023

The Easiest Part of LLM Applications is the LLM

Building LLM-powered applications involves more than just the model. Challenges include deploying and managing applications, navigating cloud resources, tracking performance, access control, credential management, and budgeting for costs. The evolving nature of LLM applications amplifies these challenges, requiring a focus on data, cloud infrastructure, and cost management for successful development.

Hide Summary

Substack

February 19, 2024

Needs Before Tools: A Pragmatic Approach to AI Workflow Integration

The article titled "Needs Before Tools: A Pragmatic Approach to AI Workflow Integration" emphasizes the importance of prioritizing needs over tools in integrating AI workflows. It provides insights on how to approach AI implementation effectively. The article is currently in draft status and was created on February 20, 2024, with the last edit made on February 21, 2024. For more details, you can read the full article at the following link: https://tfthacker.substack.com/p/needs-before-tools [https://tfthacker.substack.com/p/needs-before-tools].

Hide Summary

Noemamag.com [http://noemamag.com/]

October 10, 2023

You're probably going to use Microsoft Copilot wrong

Advanced AI models, such as ChatGPT and others, have notable flaws but already demonstrate essential features of Artificial General Intelligence (AGI). These "frontier models" can handle various tasks, languages, data modalities, and learn in-context. Reluctance to acknowledge AGI often stems from skepticism about metrics, alternative AI theories, human exceptionalism, and economic implications.

Hide Summary

Simon Willison’s Weblog

November 15, 2023

Exploring GPTs: ChatGPT in a trench coat?

This document explores various aspects of GPTs. It discusses the use of GPTs in generating unique recipes with animal chef personas, the potential of actions and the datasette.io [http://datasette.io/] database, the limitations of the knowledge feature, the GPT Builder's functionality, the combination of GPTs with other modes, the billing model, prompt security, and future wishlist items such as better documentation, API access, and making GPTs available to non-subscribers.

Hide Summary

Lillian Weng

June 23, 2023

LLM Powered Autonomous Agents

Building agents with LLM (large language model) as its core controller is a cool concept. Several proof-of-concepts demos, such as AutoGPT, GPT-Engineer and BabyAGI, serve as inspiring examples. The potentiality of LLM extends beyond generating well-written copies, stories, essays and programs; it can be framed as a powerful general problem solver. Agent System Overview In a LLM-powered autonomous agent system, LLM functions as the agent’s brain, complemented by several key components:

Hide Summary