Skip to content

Latest commit

 

History

History
49 lines (41 loc) · 3.35 KB

vision.md

File metadata and controls

49 lines (41 loc) · 3.35 KB
Raw

This paper describes our vision, you can also check out our:

Gadfly

Imagine a world where every person can generate and interface with computer programs at the speed of thought. You think about an encoding or representation, think about the kinds of high-level information you want to know, and an AI running in the cloud generates and runs an implementation for you. It probably uses data from your phone or contacts or whatever, it can see and hear what you see and hear.

Notes

The vision

  • 1980 people who wanted a computation done had to go to a programmer

  • Mathematica/Wolfram Alpha

  • linguistic interfaces broaden access to computation

  • slabs of boilerplate become a single function

  • boilerplate programming goes the way of assembly

  • natural language -> computational language

  • vastly more people will care about computation. art teachers will use computation more

  • "why does the picture look kind of off"

  • math (learn it before you can use it) compu lang (use it then learn it)

  • will people never leanr compu lang

  • what should people learn (cx computational thinking)

  • lots of fields were became computational (software eng uni starting in the 90s)

  • you need know where to drive the car (encoding/decoding ideas)

  • cx is a way of formalizing the world, kind of like what logic was back in the day. the growth of systematic data (systematic description of certain things (encodings)) a formal way so that you can build up

  • we implement it with computers but what it really is a formalism. it is constrained but it can expand to cover vast spaces

  • will language evolve into computational language?

  • language is optimized for shallow context, it doesn't do a bunch of deep nesting because our brains don't do that well. interestingly, chatgpt is good and bad at the same things (it doesn't do paren matchign well)

  • formalism is sometimes math, sometimes programming, etc

  • learn about encodings

  • the computationalization of the world should be part of standard education

  • testing is part of it, it's not just encodings

  • analysis, breakdown, reduction, expansion, etc

  • it's kind of a set of design patterns that end up showing up everywhere as you write computer programs. it's the kind of computational structure of the world. not the theoretical, the actual in fact practical logical structure of the world

  • writing started to be a specialty, then it became a thing for everyone, math started to be a specialty, then it became a thing, programming started to be a specialty, then it became a thing. we had quills and parchment and bullshit, but then it became accessible. we had compilers and programs and whatnot but now we have language models. we will expect everybody to know how to do cx because it will be available to everybody

  • the visual arts will be able to do something like "tell me what % of the screen is white".

  • GeoHot -- why don't LLMs have a memory or hippocampus?

  • future LLMs will be smaller, but loop and have retrieval (so you can cite sources)

  • when someone makes an LLM that can cite its sources it will kill google

  • competitors (Meta is better than Google)

  • the "do what i mean machine"