← 3c PAL


TOC

  • 3c.0.1 The PAL framework simplifies AI app tasks
  • 3c.0.2 PAL (no app AI)
  • 3c.0.3 PAL + app AI


3c.0.1 The PAL framework simplifies AI app tasks

Real world apps change quickly. PAL apps are used to analyze complex environments that are constantly changing. You need a framework to allow focus on the parts that matter. PAL is a framework that manages the complexity of creating AI apps.

drones

I attempted to show how to create app agents (AI, non-AI) without a framework in 3.1 code-first agents (I did not really understand how PAL worked then). Its a lot of work. Complexity. Its difficult to focus on the core task.

drones


3c.0.2 PAL (no app AI)

This was the original PAL architecture (25 years ago, shortly after 9/11). This is what demos D1-D7 (below on this page) are all about. Detecting situations without AI.

drones

I did some demos to test this kind of functionality without a framework on page 3.3 Agent PAL core demos (again, my knowledge of PAL back then was limited to what I read).

drones


3c.0.3 PAL + app AI

I started adding AI to the end user app in demo D8.

drones

There are 2 problems when using AI for detection

  • (1) Data sources are often messy. You need something like PAL pipelines and ontology to clean things up.
  • (2) Thats important because AI is just pattern recognition. On the (rather short) page Hack I layout the gist of what an AI LLM transformer (TF) does (which shows the core of AI really is). Most importantly on that page I show that LLM TF’s are simply pattern classification algorithms. The classification of input tokens determines the output tokens (TF’s understand nothing about language, they only generate a classification number for the input numbers). Recent GPU performance breakthroughs mean the patterns can scale to massive sizes, increasing the sheer number of patterns that can be reliably classified. That is not intelligence, but pattern matching. An amazing hack. And it works for the most part. But it needs guardrails.

That’s where PAL comes in. PAL minimizes the limitations of the AI hack while maximizing its effectiveness.

The LLM transformer is a hack, but Palantir provides the guard rails that maximize its effectiveness
drones

I did demos to test this kind of functionality without a framework and with AI in the app on page 3.3 Agent PAL core demos.

drones


26.0725 (v1 26.0720)