Ideas and Notes towards considering OpenAI-GPT3 as a network of HyperObjects (Expressive Maps)

gpt3ashyperobject.md

GPT3+ as HyperObject

Mindmap of the Idea

https://www.xmind.net/m/YScgEU/

Markdown/Outline of it

Human Language

What It Does

• Common Notions

  • Symbolic

  • Stories/Narratives

  • Efficient

  • Universal Grammars, blah blah blah

    • Go read all the Pinker, Chomsky, etc
    • Or consider Skinner
    • Or consider Fodor etc

• More General

  • Maps

    • Maps between Humans
    • Maps between Computers
    • Maps between Computers and Humans
    • Maps between Self

  • Computes

    • Describes Itself
    • Mutates
    • Determines
    • Undetermines
    • Reasons
    • Points To and References Other Things

  • Space and Time

    • Maps Causality

What is Language?

• A Multi-way System

• Metaphors and Analogs

• Designators

  • Rigid
  • Loose
  • Determiners

What is GPT3?

A Hybrid Human-Computer "Natural Programming Language"

• Ideas/Premise

  • Consider it COMPUTATIONALLY

    • We know it's irreducible.

  • Worry Less About Grammar and Human Language

  • Worry more about Computational Expressivity

    • What new things can we say, do, paint, compute?

Alien Language

• Grammar is Human Adjacent

• Non-Deterministic Programming Language

• Byte Pair Encoding

  • Tokenization
  • A map between human languages
  • Including our "markup" and "scripting" language

Address Space/Map

• Maps Language to Itself in a HyperGraph

  • Language already does this, but not in this "predictive" feed forward, recurse back way.

    • Humans do this, but computers hadn't done it yet

• Each Prompt/Seed and the parameters are an address to a possibility graph

A Compression, Encoder/Decoder

• Compresses the Web (and whatever else we throw at it)

  • Lossy, most likely
  • Losses some of the non lingustic structure
  • Gains linguistic structure
  • Decoding Expands beyond the original compression

The HyperObject

Uses/Human Extensions

• Avatars

  • Self Version Generators
  • Vanity Computing

• Co-Programming

• Co-Writing

• Co Music Making

• Co Conversation

• Co-Learning

  • Co-Teaching

Universal Object Locator

• HyperObjectLink: https://gpt3.openai.com/modelsnapshot/prompt/temperature/top_p/n

  • model snapshot
  • prompt
  • temperature
  • top_p
  • n (choices/alternatives)

• Registration

  • Properties

    • Intended Use

    • Age Restriction

    • Abstract Description

    • Curator/Finder/Prompter

    • Restrictions on Use

      • Licenses

    • Expires

  • Commercialization/Monetization/Community Viability

    • Price Per Use
    • Blockchain or other way to do smart /on the fly contracts
    • "Patreon" Type model for hunting for novel HyperObjects
    • Wiki media type model

• Registration Server

  • Community maintained HyperObject Name Server (HNS)
  • Discoverable HyperObjects by Property
  • Consider joining up with WikiData/Wikipedia to extend those efforts

• HyperObject SDKs

  • By Programming Language

  • Simple Libraries to make compsition of HyperObjects very easy

  • Non Lingustic UXs

    • Programmatics
    • Mobile/Desktop Visualization
    • Template Engines
    • Bots/Avatars
    • Robotics

  • Interops/Extensions

    • Wolfram Language/Wolfram Alpha
    • Google Knowledge Graph
    • Social Graph
    • MusicBrainz etc

• HyperObject Peer to Peer Network

  • CLient

  • Protocol

  • Distributed Database/Middleware

    • The idea is that you can have collections of HyperObjects that do work together

GPT3 Core

The Models/Engines

• Can we mirror these?

Infrastructure/Network/Hardware

• Is there a way to federate this to more than OpenAI?

  • Cost
  • Infrastructure
  • Cycles

Engine APIs

• Can we mirror these?

Inspirations/Precursors

Wolfram Function Repository

PIP/PECL/PERL.org/NPM/GEMS

WikiData

Semantic Web/RDF

HTML/HTTP/WWW/W3C

Old Logarithm Books

• Engineering Handbooks