TOC

  • 0 How I started learning Foundry
  • 1 How I am programming (and documenting) the ZAI Foundry Quick Start
  • 2 This is the future of programming Foundry (and writing documentation)
  • 3 Such changes are the norm



0 How I started learning Foundry

I spent a couple of months going through Foundry demos (see “3.2 PAL project/doc sources”). They were super helpful. But what I needed was

  • A personal trainer, an expert with Foundry.
  • This was critical for fixing unexpected problems.
  • Feedback on the accuracy of what I was doing.
  • Ability to change the demo focus, direction, level of complexity, etc.



1 How I am programming (and documenting) the ZAI Foundry Quick Start

I am doing the Foundry programming for the 2.1 ZAI start-to-finish demo project with the invaluable assistance of FDE/GPT.

For documentation (docx #695):

  • Most of the text is from FDE chats.
  • Some of the screenshots were submitted to FDE for feedback (I think GPT provides the “vision” and FDE the direct link to my Foundry project).
  • What the reader probably needs most are what prompts to ask FDE. The most important part of learning Foundry is learning how to use FDE or GPT (AI) to help you learn Foundry.

The new programming/documentation paradigm with FDE/GPT.
drones



2 This is the future of programming Foundry (and writing documentation)

FDE does all of this with NO intelligence. It does not need intelligence (humans supply that). An improvement in FDE performance only requires better training and more powerful/efficient GPUs.

In the (near) future

  • FDE can directly “see” your Foundry instance dialogs (which has tight security controls, or you use an external camera that you can physically control for extra security).
  • To interact with FDE, you say (audio) what to do, point at the screen, or you simply look at the area of interest.
  • FDE makes the required changes directly.
  • The bar for programming productivity will be raised once again.

This will require massive levels of computing power:

  • Efficient GPUs (a tech level monopoly like Nvidia has now)
  • Commodity LLMs (the LLMs will be secondary in importance; already the choice of LLM for FDE is not that important; we are not driving a car, just programming)



3 Such changes are the norm

A few example of such changes

  • Such programming paradigm shifts have happened before.
    • I remember the first webpages. They were incredibly difficult to use, and most of us did not see much value in them. We were quite mistaken.
    • The tools eventually arrived for the non-expert to create websites. Its impossible to imagine life now without them.
  • I previously worked on a project that was a failed attempt at creating such documentation. 25 years ago datango, a Berlin startup, had a product that was trying to do what I am doing now using FDE/GPT to help me program. datango’s SW created popups inside a browser with instructions. But this was not AI, had to be programmed, and the startup idea failed (not sure what happened to the startup).
  • An experienced pilot talks about how he believes airliners are already “basically” flying themselves.
  • I’ve read about the incredibly fast turn-around for manufacturing from Anduril and others (probably with heavy usage of AI in a support role).
    The new Anduril factory in Ohio (source).
    drones drones