Photoshop for Text (2022)

Oh my god. I wrote a whole library called "Constrained Text Generation Studio" where I mused that I wanted a "Photoshop for Text". I'm not even sure which work predates the other: https://github.com/Hellisotherpeople/Constrained-Text-Genera...

The core idea of a "photoshop for text", specifically a word processor made for prosumers supporting GenAI first class (i.e oobabooga but actually good) - is worth so much. If you're a VC reading this, chances are I want to talk to you to actually execute on the idea from the OP

I love these ideas as I have found ChatGPT immensely helpful for refining my thoughts into polished text. It's like a brainstorming partner that helps streamline my ideas into something coherent, enhancing the flow of my initial drafts.

I'm intrigued by the concept of a Google Docs-like interface with LLM-powered editing tools for tasks like sentence merging or idea integration, essentially a text version of content-aware fill.

Currently, I'm manually doing what I envision could be automated. Even better if it's built into the OS, enhancing every textbox with such capabilities natively. Like the original post suggests, a local solution would be ideal. I'm hopeful for a future where Apple integrates this into macOS with local models.

I suspect that there will be substantial limitations on reliability for quite some time. Photoshop has the same limitation; but problems are usually fairly easy to spot. Vetting LLMs’ output with high reliability, however, would take nearly as long as editing the text by hand. That isn’t to deny that they could be quite useful when the stakes are lower.

Extendable text-editors like emacs already are starting to see integration with local language-models. What the article describes sound like quite straightforward functions to add once you have some plumbing in place to send text selection to a LLM, unless I am missing something?

Gotta love reading ostensibly technical articles in the AI age.

1. Oh, this is an interesting problem

2. Asserts that it is possible to implement

3. How will it be implemented?

4. “I don’t know how but soon AI will”[1]

5. (Bonus) Recurrence of author’s own term that he “coined” in a one-minute read blog article

What is even the point? I too want stuff, vaguely. I can’t implement any of it though. What about this compiler optimization that I want? I could write an article about it because soon-AI.

There’s already plenty of interesting things you can say on this topic without invoking spooky LLMs.

1. Transforming abbreviations/initialisms/contractions back and forth

2. Transform inline numbered list to numbered list block

3. Switch between parenthetical remark and footnote remark

I don’t know, maybe those things aren’t interesting. Maybe just pedestrian. But it seems easy enough to support in a structured format and I’ve never seen it before. Why not?

Maybe I’m too Programming 1.0. I like conceptual problems that can be grasped. I don’t like black box technology.

[1] For some reason I can’t select the text in the article. The text… and I can’t be bothered to copy from the source.

It had better include the "Pierre Menard" tool, which produces a symbol-for-symbol copy of the input text, which, however, is much richer in allusion.

I'm not sure I buy this. I love using AI to help with coding and other questions, but often times I feel like AI is too agreeable to be an effective writing partner.