[WillAdams]

To some degree, that's what one had w/ Apple's Newton Intelligence on the MessagePad --- it was "just" fancy pattern-matching, but mostly it worked, and the UI and implementation were quite good, and it kept me organized all through college.

[JKCalhoun]

Mentioning the Newton may be anathema to the discussion (it seems to bring up the usual jokes, etc.) but I was thinking too that the Macintosh (or the Xerox Alto if you like, or the Mother of All Demos) tried to move us in that direction by "skeuomorphising" the computer interface—make it look like the more familiar "real world". The Newton pushed further. It seems to have been on the mind of at least a few people at Apple.

It sounds like the author is on the same track, has the same mindset. And I like.

I am also reminded of the Young Lady's Illustrated Primer: in Neil Stephenson's Diamond Age. It is not exactly what the author describes but, if the book had a computer backend, it also divorces the user from the computer interface we have come to know. Perhaps for me some future (better) local LLM within such a book is what I want. A kind of companion I ask questions of…

(I mean I suppose I should just do what was posted a day or to ago to the Ask HN: and put a local LLM behind a messaging app and I could just converse with it wherever I am. Tangent: I am kind of fascinated by the idea of a personal LLM that has context stretching back to my earliest days—were I to have started conversing with this synthetic companion at a young age. Imagine the lifetime of context where the LLM knows my habits, how I've changed over the years. I suppose this is nightmare fuel for a number of you.)

[lamasery]

Other copies of the Primer do have a computer backend.

There are basically three versions of the book:

1) The ones developed for a few rich kids. These are partially automated, but backed by gig workers. They get what we might call (if you'll pardon the term) "Actually Indians" AI (augmented by the regular type).

2) The one our protagonist gets. This is one of the books from #1, but the distinctive feature here is that an early gig worker (the book calls these "'ractors" when they're doing this kind of work) the protagonist draws takes a special interest in her and intentionally keeps drawing jobs for her over a period of several years. This continuity and personal care by a single real person is what sets it apart and makes her experience so excellent.

3) The mass-market version that's entirely computerized, no human touch. This version brainwashes a fuckload of kids into becoming the "mouse army", and that's really all we see as far as what it can do: something really bad (if convenient for our protagonist).

The message of the book is 100% the opposite of "automated learning-books are amazing". It's "tech for learning sucks ass and/or is outright dangerous if you rely only on it, and a real human tutor who cares about a kid is the best thing around even in a crazy high-tech future-world".

[WillAdams]

Charles de Lint had an intelligent book in his fantasy novel _Jack the Giant Killer_ (or maybe its sequel) --- I've tried doing the conversing/chatting thing w/ an LLM a couple of times, but always got annoyed more than amused.

What's the point? LLMs tend towards the mean/average --- I want better in my life and interactions --- it's useful when I need an example DXF or similar rote task, but my current project is a woodworking joint which has no precedent.

Yes, the skeumorphism angle is an interesting one, and one which is surprisingly absent in the _ur_ description of a stylus equipped computing device, the slates/tablets from Larry Niven and Jerry Pournelle's _The Mote in God's Eye_ --- this sort of thing seems to be coming back around --- a recent Kindle Scribe firmware update add shape recognition. I'd be _very_ pleased if my new Kindle Scribe Coloursoft could fully become a replacement for my Newton....

[JKCalhoun]

I think you're right that the use case for an LLM is still rather niche. It's perhaps still worth exploring though as they may well improve over time.

Regardless, I have still found them useful. Diagnosing the problems with a car is maybe an esoteric example but is still useful.

For many months now I have been working through learning about and implementing a hobbyist analog computer with LLM as engineer-confidant. I already knew the basics of op-amps and analog computing but was surprised at a lot of the new things I discovered only by way of the LLM saying (for example), "Hey, here's a nice way to get your reference voltages…" and the project benefited from it (and I learned about a new chip/device/technique).

[WillAdams]

Yes, they do work well as a stand-in for the "competent technician with skill in the pertinent art and and fully aware of all prior art" (to use wording like to the patent application standard).

But it's only going to allow you to avail oneself of prior art/techniques.