I love this piece by @nelmqvist.bsky.social on LLMs. This gets to the heart of why code opened my mind, years ago. I learned how to communicate using precise, unambiguous terms. Writing code taught me a new way of thinking, which was life-changing. niklaselmqvist.medium.com/the-natural-... - ThreadSky

frank.computer • 52 days ago

I love this piece by @nelmqvist.bsky.social on LLMs.

This gets to the heart of why code opened my mind, years ago. I learned how to communicate using precise, unambiguous terms. Writing code taught me a new way of thinking, which was life-changing.

https://niklaselmqvist.medium.com/the-natural-language-bus-fb083c293851

Comments

frank.computer•52 days ago

Niklas's metaphor of a swiss army knife versus a surgical tool is brilliant. As someone who considers themselves a tool-maker, I don't want to make everything into a swiss army knife. I want to develop unambiguous, specific tools that help people form new mental models and do new things.

frank.computer•52 days ago

I contrast this to @infowetrust.com's recent piece on "A Dangling Blue" where he *doesn't* want to think about domain-specific details and instead just wants to drive a car.

"The systems we use to make charts are in service to the machine, not to us."

https://www.chartography.net/p/a-dangling-blue?trk=feed_main-feed-card_feed-article-content

frank.computer•52 days ago

There's a tension here!

Niklas is arguing for domain-specific tools and languages. He is making the argument that it is *good* to learn how to work with unambiguous tools and to learn new languages.

But RJ is arguing that this all takes too long, it is too painful. We need to go faster.

frank.computer•52 days ago

RJ's car metaphor is interesting because virtually all of my time over the last decade that I have been visualizing data has been in service of others: I craft tools to help others build, I prep data and push pixels to help others see, etc.

frank.computer•52 days ago

I'm the one who thinks about how the engine works precisely because I believe in what RJ is arguing: some people need to go faster, need their equipment to be easier to use, and have a lot more left to do.

I want to help them get there.

frank.computer•52 days ago

But can someone drive without learning the "language" of a car? Won't there *always* be an effort involved in learning something new? We need to learn how to start an engine, what maintenance is required, what the rules of the road are, and how to navigate your way through difficult terrain.

nelmqvist.bsky.social•52 days ago

Absolutely. Thanks for highlighting this post. Many people have already observed that gold rush LLM papers are often thin interface veneers on top of ChatGPT. I think it is intensely unsatisfying to give up all that we have learned of specification and interactive UIs for imprecise natural language.

frank.computer•51 days ago

I agree! I feel like the LLM alarm goes off, I wake up to see another wrapper, and then I just spam my snooze button again and go back to sleep. I want to wake up when it gets good, but otherwise I've got more interesting dreams to get to.

duvekot.bsky.social•51 days ago

Mastery is the cost of expressiveness. If that was easy to solve, we’d have invented a (better?) violin that people like me could play. There isn’t one: We invented record players instead.

frank.computer•51 days ago

Yeah, I really do believe that some tooling simply cannot be made more concise or easier without some loss of expressiveness, error, ambiguity, or control.

Comments

Posting Rules

Reply