Exploring user editable interfaces
Australia
Joined December 2024
- Tweets1.9K
- Following368
- Followers143
- Likes24.7K
I wonder whether you could build a semi-offline internet:
1. Google search index - around 5GB including text of landing pages, synced monthly
2. Social media - thousands of posts and text of landing pages for any links, synced daily
3. Wikipedia
4. Local LLM
Internet apps function normally, local proxy provides local cache results. Anything not in cache gets added to next sync + spider.
AI rant:
If I were to start a company, I would employ people who I know could work independently. I would give them some tasks, there could be some back and forth and feedback, but generally speaking I've employed them because I trust what they can do and I'm basically going to accept whatever they produce, even if that could be quite different to what someone else would do. I don't have to micro-manage their output even though their output may be wildly different to someone else's.
An example would be Apple's Lisa and Mac projects. Hearing stories about how those products were developed you rarely hear about Steve Jobs. Those products could've taken wildly different paths. Jobs may have provided some input here and there, but he was trusting his employees to produce whatever they thought was suitable.
What I want is one of two extremes:
1. A machine, that I can push and prod and provides deterministic outcomes
2. A human, where I will mostly accept whatever they produce
LLMs live between these two worlds. They aren't deterministic precision machines and they aren't humans that you can mostly accept whatever they produce.
Currently the two approaches seem to be:
1) Treat the LLM like a machine. A lot of specification upfront, huge harnesses, and loops to try to get a fuzzy neural mesh to operate like a precision deterministic machine.
2) Just trust the LLM like you would a human. Accept whatever slop it produces, with some mild input, and ignore the haters.
Personally I think LLMs will become more and more human-like, ultimately indistinguishable from humans, and the trend will be increasingly to take the second approach.
An interesting movement could be, wow look what an LLM can do, now let's try to do it without an LLM.
I've wondered now that we know which parts of LLMs are useful, how much of that functionality we could build without using LLMs.
As a parallel, we got Doom working on a C64. We didn't actually need a 486.
For example, LLMs work well as an intelligent Wikipedia, but how close could we get to that functionality, by changing the way we interact with Wikipedia?
Whenever I see something AI generated now I see it as a tree with many branches and every leaf is a dead end.
User Editable retweeted
Anyone else get told by developers that AI can replace designers, then get asked for a Figma mockup showing what a disabled button should look like?
User Editable retweeted
When was the last time you designed something without using AI at any point in the process?
User Editable retweeted
Doing an implementation of Sam and Ravi's ragged blocks in Hazel to help visualize some stuff
Ragged Blocks: Rendering Structured Text with Style by Sam Cohen
arxiv.org/abs/2507.06460
very interesting method for rendering nested structure while respecting typographical layout
User Editable retweeted
Claude Code does not work well for perfectionists like me who often polish graphics pixel by pixel.
I would spend 20x more time trying to fix its mistakes creating a photorealistic bookshelf instead of just doing it myself manually and visually in Framer.
What's the likelihood key technologies would never have existed if the key people behind them never built them, e.g.:
- LISP: John McCarthy
- Smalltalk: Alan Kay
- Erlang: Joe Armstrong
The moat is you.
User Editable retweeted
Everyone’s seen the Sketchpad demo. Few understand how it actually works.
I reimplemented Sketchpad from scratch, and I’m making short videos that recreate Sutherland’s demos and explain how they work. The first 8 are up — from drawing a line to the famous rivet.
Link below 👇
Some people need to hear this:
It’s okay not to use AI.
(It’s also ok to use AI, but I don’t think people need to hear that at the moment)
User Editable retweeted
Some random things I need to say out loud:
1) AI so far is dangerously converging power in the hands of few and increasingly swapping labor for capital/compute
2) In a hyper capitalist environment that’s a recipe for disaster without rethinking income and what gives life purpose
3) Some of you are way too eager to give up on your own thinking
4) Some people and ideas are a cancer to democracy and society. Support is a choice. So is not associating.
5) We are at an all time low for ethics, principles and morality
6) Even thinking of AI welfare before welfare of all humans is deeply immoral
7) New Siri still sucks
8) I don’t quite get the iPhone Duo
9) The Mission S is beautiful
If the world's top A/A+ coders said they no longer want to use AI, the top companies would have to settle for B/C coders using AI.
I suspect low velocity A+ non-AI software would be better than B/C AI software.