Today and yesterday I let Claude, Gemini and ChatGPT improve my Notes setup on my blog. (I don’t pay for any of them, so I cycle through them.) Last week, I had this epiphany that I have a lot of content I wrote out there that lives not on my website, but other platforms. Lots of tweets, yes — but also tiny blog posts I wrote years ago.

I call them „notes“.

I wanted to bring them all together: For archiving purposes, but also as a backlog for creating more of them. So I added all posts from my TheLisaProject Tumblr, LisaNY Tumblr, and the Notion „blog“ in which I document all my Color book updates. I was positively surprised about both the Tumblr and the Notion export, and how quickly the LLMs came up with a working Python script that turned JSONs and HTML files into Jekyll-compatible Markdown files with a consistent frontmatter. Neat!

Today and yesterday I finetuned everything (e.g. by fixing Markdown formatting or setting up an RSS feed for these notes), and I set up a GitHub Action that checks every six hours if there’s a new post on my Tumblr and then takes its content and puts it in my GitHub page.

Meaning: I’m writing these lines in a convenient Tumblr app, but within six hours, they (and the photo on top) will live in my own blog. That’s amazing. I’m very happy with it. And I couldn’t have done it within days without those LLMs. It’s one of these projects that I, as non-programmer, wouldn’t have started if LLMs didn’t exist.

My son has this great crayons. They make it super easy to draw big areas quickly…and make it impossible to draw in a precise way. Indeed, the pencil sharpener for these crayons doesn’t let you truly sharpen them. They’re like „Nope, there’s no way you’ll draw a precise line with these crayons. We won’t let you. The world is messy and best captured with imperfect strokes. Deal with it.“

This, my friends, is a crazy tool.

I leave you with it while I’m out on vacation next week. I’ll be back on Monday, October 28th 👋

Enjoy the sunny fall (or spring) days!

Very happy with the process I made today on the blog post. Turns out, ChatGPT isn’t that bad in coming up with color palettes. A prompt like “Give me four colors that will fit to #FF7651 and #006C67 for a categorical palette for my data visualizations. All colors should have the same visual importance.” works fairly well.

What ChatGPT isn’t doing well is creating color swatches, as you can see on the top…

Well, I’m sure it’ll be great at that in a year from now.

Small advances regarding the blog post. Baby steps, baby steps.