In 2022, ChatGPT introduced the general public to the power of prediction through Artificial Intelligence, specifically Large Language Models (LLMs) and Generative Pre-trained Transformers (GPTs). Overnight, it became possible to use natural language to get instant answers to nuanced questions and generate original content, including text, images, video and audio - as if by magic. Yet after 18 months and $60 Billion of investment, the use of LLMs and GPTs in most end user products and services remains superficial, inefficient, and proprietary.

This series explores how we can unlock AI technology's full potential for end user computing for both consumer and professional applications. It draws on research from economics, psychology and software engineering, as well as from experience architecting AI friendly systems and leading product and engineering teams through implementation.

Contents:

1. The value AI provides

2. Learning from the past: Steam > Electricity

3. Problems holding AI back

4. What AI first applications could be

5. Industry Alignment & Recommendations

Works Cited:

Prediction Machines

Power and Prediction

The value of AI

The term AI covers a range of technologies, including consumer tools like ChatGPT, fraud detection in finance, and medical imaging analysis in healthcare. While these applications seem unrelated, in each case AI is really doing the same thing: prediction.

Part 1: The value of AI

Part 2:

by part 1: distilling the value it creates, part II: architecting a system that can take full advantage of it, and finally part III: outline the biggest problems and propose solutions.

imagine what the user experience could be in a world designed for AI.,

and reduce AI to the value it creates, then reimagines a world designed to take advantage of that value.

To unlock AI's full potential, we'll need to reduce the value it creates through the lens of economics, technology, and society.

In this series of posts, we'll reduce AI into specific

Fist post aims to unpack the fundamental benefits that toady's AI, LLMs and GPTs provide, through the lens of economics and history

seeks to explore

Like Steam Power, Electricity, and the Internet, Artificial Intelligence is a General Purpose Technology,

In 2007, the web began to fundamentally shift from favoring 'mice and monitors' to today's 'touch and mobile' experiences, as the primary interaction moved to the smartphone. Now, in 2024, we are on the verge of a new chapter, one I predict to favor 'chat and widget'.

The key to unlocking the value of AI is to decouple predictions from judgement, ideally in ways that let us minimize humans getting in the way. Instead, we'll make requests using natural language, empower the AIs to predict the desired result or actions to take, then let us confirm it. This pattern requires two critical changes to how we use the web:

1. From Browser to Assistant

   1. Local - on device - agent

   2. Has access and context, our preferences

2. From Search Engine to Assistant

To better leverage the benefit of better predictions in how we develop experiences on the web, we need to make the web a more semantic place.

Problems

• Privacy / Personal Data Context needed

• Lack of Semantic, Machine Readable 'App Store'

• Most actions/mutations/useful functionality is behind proprietary interfaces

  • available actions

  • identity (sso, payment)

• Most experiences are only exposed as whole pages

System Changes Required

1. Make the web semantic, explicit, and navigable for on device intelligences

   1. Better content alternative formats that dont require full scraping in order to index

      1. Schema.org

   2. Expose a 'sitemap' for APIs / Tools offered by a domain as machine readable openapi schemas published at a well known discovery endpoint

      1. Need search engines who index these services and