Geoffrey Litt Profile picture
May 4 9 tweets 3 min read Twitter logo Read on Twitter
I wanted to convert a JSON file of a chat transcript into nice markdown text for sharing w/ people...

so I had GPT generate an ephemeral React UI where I can drag in the JSON file and it outputs the markdown🤓

reflections on the process:
My first idea was to just have GPT do the conversion itself: paste the JSON in and tell it to output markdown.

Worked fine, but slow, might not work on long transcripts, and hard to know if it's 100% reliable. Image
Next idea: have GPT generate code to do the task!

Just told it to make a node.js script that outputs "a nice markdown file", didn't bother specifying formatting details

Worked fine on first try. but! annoying to invoke a script in terminal, futz w filenames, etc Image
Then I realized I really just wanted a GUI where I could drag in the JSON file directly and then copy out the markdown result.

Asked GPT to turn the node.js script into a react app: Image
then I just copy-pasted the react code into @Replit and it worked on the first try :)

replit.com/@GeoffreyLitt/… Image
this is a terribly simple UI, really just a place to drag a file and run a simple data conversion.

and yet -- I wouldn't have bothered to code it manually. only worth it because it literally took 10 seconds to code both the data conversion and the GUI
It’s also neat that I now have this GUI tool and can quickly reach for it whenever I need.

I think ChatGPT w/ plugins could have produced a similar UX for a single conversion, but wouldn’t have left me with a solid reusable tool
Also, interestingly I haven’t even bothered to read the code.

Just checked a couple example outputs which was enough to give me confidence that it does the right thing
Just a tiny, real, useful example of a pattern I’m excited about: personal AI-generated software tools

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Geoffrey Litt

Geoffrey Litt Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @geoffreylitt

May 3
This video is a fun case study in chatbot i/o modalities.

It claims that ChatGPT-over-voice is better than Siri, but I actually prefer Siri for this example…
Siri: responds immediately with a written sentence that I can read in the blink of an eye. Efficient.

ChatGPT: takes a while, slowly speaks out loud the answer
Takeaway for me: if I’m engaged with the task and can look at a screen, voice input + visual output is a nice efficient combo. Voice input is faster than typing but don’t necessarily need voice *output*
Read 5 tweets
Mar 19
What if -- despite all the hype -- we are in fact underestimating the effect LLMs will have on the nature of software distribution and end-user programming? some early, v tentative thoughts: 1/
it seems likely to me that all computer users will soon have ability to 1) develop small software tools from scratch, 2) describe *changes* they'd like made to existing software they're already using. what will this mean for software ecosystems?? 2/
you, skeptic: "nooo but the LLM-generated software will be lower-quality than handcoded software by pro teams: filled with bugs, uglier, bad... 😡 "

Yeah totally true... also all qualities which also apply to spreadsheets vs "real apps"! 3/

Read 22 tweets
Mar 17
If you want to think crisply about live programming interfaces, highly recommend this paper by @qualmist and @jeffrey_heer!
We already talk a lot about "liveness" (fast visible feedback) in programming, but this paper argues there's another quality we want: *domain-specific* UIs for editing programs.

They call this quality "richness". Image
I've already found this to be a useful lens to apply in my work. Often a programming system that "seems vaguely cool" is only live or only rich, and if you have the vocabulary, you can be more attuned to possibilities of going further along one axis...
Read 4 tweets
Nov 7, 2022
1/ "Build an app with a spreadsheet" is a mainstream idea now (Airtable, Glide, Coda, Notion, etc) but I believe there's still so much room to push this model further.

Quick list of some (lesser-known??) projects that I find thought-provoking in this space:
2/ We've seen "project tracker" and "inventory list" a bunch, but what's the ceiling on building complex software w/ spreadsheets?

Here's Alan Kay speculating in 1984 (!) about how you could use a spreadsheet and a drawing app to build a word processor

worrydream.com/refs/Kay%20-%2… Image
3/ The details of his explanation get intricate and I'm not sure if they'd work, but I like his motivation. The idea isn't that everyone will build their own word processor. It's that when they "open the hood" of their existing tools, they have a chance of understanding/tweaking Image
Read 18 tweets
Nov 1, 2022
✨ Excited to share Potluck, a new project I worked on w/ @mschoening @_paulshen and @paulsonnentag at @inkandswitch!

Guiding question: what if you could start with a text note and gradually evolve it into an interactive software tool? 1/

inkandswitch.com/potluck/
Consider jotting down a recipe. Where do you put it?

Could put it in a notes app: simple, fast, low-ceremony.

Or put it in a "proper" recipe app. Gives you nice features like scaling up the ingredients, but also more rigid and formal.

Why can't we have both?? 2/
Potluck is a research prototype where you start with a text note, and gradually turn it into an interactive tool. You only interpret structure and add computation when you need it.

Eg: write down a coffee recipe as text, then scale it up computationally 3/
Read 13 tweets
Mar 21, 2022
Interesting looking at the docs for NeXT WebObjects and revisiting their rationale for "why you need an ORM layer instead of just putting your biz logic directly in the database"...
1) "Databases are slow so you need an ORM"

Makes sense in an architecture where DB queries require network roundtrip. But not true if your DB is colocated w/ your logic / is fast!
2) "Your logic is now portable across databases"

This one seems kinda legit, although portability is dicey in practice. Even migrating between "postgres-wire-compatible" DBs is not trivial
Read 5 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us on Twitter!

:(