🧡

See how I let LLMs rank Zapier's top 10 AI automation articles. @zapier @wadefoster https://t.co/uCQH70yC5m

Do you ever feel misunderstood by AI models? I do. When that happens, I usually think, "You're so smart. So why don't you get it?" I realized it tends to occur when I'm not being clear. 👉

What instruction do you use to tell LLMs to be concise? I often add something like [12 words max] or [12 words max for each point], but there might be a better way to do it. And you?

0) Know what you want.

What do you actually want from a tool? 1) Solve my problem. 2) Be convenient. 3) Integrate with my other tools. 4) Be extensible, or simple to replace. What else?

It really feels like Emacs but for tabular data, with a more mainstream scripting language. Give it a shot and tell me what you think:

It's extensible: put some Python in your .visidatarc file and you're done. It's self-discoverable: you can inspect the current options, commands, bindings, and built-in guides: - Shift+O - gzCtrl+h - Ctrl+h to enter the help menu > Visidata feature guides

I love the Visidata paradigm. Everything is a sheet. (Visidata is a TUI interactive tool for tabular data) You open a .tsv file, it's a sheet. You open a .db file, it's a sheet. You open a dir, it's a sheet. You pipe in processes with `ps -ax | vd -fixed`, it's a sheet...

Copy-pasting snippets is easy, with no context switching. You can pass the file as context to your preferred AI assistant. No need for extra tools to gather every tiny piece scattered across many files. Simple.

Dev docs for most projects should live in one single text file. Everyone can open it in their preferred IDE and find or replace what they need. Search on docs websites is often broken. But not in my editor, and not in yours either.

I like that when I start patching a command-line tool I use, and finally find the correct flags to use and drop the patch altogether.

https://t.co/tGZScRD9bI

I don't mind digging through issue trackers, but I'm convinced LLMs could have solved my problem: 1) if the bug I hit had been clearly documented in the docs and, 2) the library's AI assistant had actually known about the GitHub issues. Read more here 👇

You can make your app 1000x faster just by finding user interactions you can remove. A human action can't compete with a computer. This optimization trick works everywhere. Remove.

Have you noticed how "The other day I was talking to AI..." has quietly replaced "The other day I was talking to a friend..."? I can't remember when this happened. Slowly, then suddenly.

No doubt, AI speeds up exploration. But finishing still depends on taste and decision. 👉

What's the new shiny programming language? Plain English. Before, with only a few words (if, else, for, def, return, etc.) you were set to program a computer. Now, it's the wild west, you need the whole dictionary.

https://t.co/drrVIxLVCG

I joined OpenAI's Vibe Engineering with Codex event yesterday. Codex as a teammate rewriting a complex project in Rust and reviewing every OpenAI PR really caught my attention. If that sounds interesting, watch the replay and check the event resources I collected! @romainhuet

It is absolutely clear to me that large language models represent the most significant scientific breakthrough of the past fifty years. The nature of that breakthrough has far reaching implications for what is happening in science today. And I believe that the entire scientific