alltom.com,
Archive,
Currently Reading,
Have Read,
Micro.blog
Subscribe with RSS or @tom@micro.alltom.com
I think it’s funny that tapping repeatedly and long pressing are mapped to power user features because that’s what kids do first and it just wreaks havoc on computer systems as a result.
The wild thing about programming with LLMs is documenting a function’s runtime cost in big-O notation and USD. “Costs $0.50 when N is 40.” I have faster and cheaper versions of that one, but I haven’t figured out how to make them as good yet, and that choice filters all the way to the UI. 🫠
“Do I have any events this week that it looks like I forgot to add to my wife’s calendar?”
I keep finding new uses for letting LLMs generate calendar files, but today I wanted to close the loop! I want it to read and write to my calendar directly.
So to start, I provided the LLM a “tool” to access Google Calendar’s ListEvents API. Just being able to read my calendar is pretty neat:
Q: What events do I have today?
Tool use: Get today’s events
A: A bulleted list of event names and times, followed by a summary like, “Looks like it’s your usual daily routine, with a trip to the museum in the morning.”
Q: Of breakfast, lunch, and dinner, which do I already have blocked out?
Tool use: Get today’s events
A: Breakfast for the kids is on there, and the drive to the museum has a note about stopping for your breakfast, but there are no lunch or dinner events.
Q: Do the kids have something planned for every weekday this week?
Tool use: Get this week’s events
A: Yep, they have X on Monday, Y on Tuesday, …
I could get used to this…
Thing I did with AI today:
Does anyone else see ML-generated ads this specific?
My long-awaited!
Congratulations, Obsidian, on graduating to my dock!
I installed it for the first-class iCloud sync, but inline PDF display and the spaced repetition plugin have made it essential.
Baa baa, terse sheep, have you any wool?
Yes.
I think that cross traffic should stop
1,794 pages in Logseq, with 279 embedded images and 156 annotated and deep-linked PDFs… but Obsidian is so fast.
My 3yos call a table of contents the “picking page” because you use it to pick a page to read 💡
If my 3yo sees an apartment building and asks, “Why are all those houses stuck together?” then someone told them the joke, right? Is that just a normal human observation?
“Outside-in” is when you make it itself
I think about that comic book movie where this magic labyrinthian map appears in the floor for like 30 seconds and everyone memorizes it for the chase scene later, but I have to keep the tab open that tells me the order of arguments to the “map” function.
Is there macOS app that gives me a global hotkey to talk to ChatGPT in a Spotlight-like popup?
Tomorrow will be a big day, just like yesterday and today, and all too many days lately.
Thanks to the power of View Source, I’ve embedded WebPPL on a page of my own site!
Just a simple (but perhaps more evocative than usual) intro to the language for now: http://alltom.com/pages/webppl-intro/
What native macOS (textual) code editors still have some wood behind them? Sublime Text? Nova? TextMate? BBEdit? I’ll try anything if someone can vouch for its working effortlessly on Sonoma.
If you’re wondering about day 1 using HEY Calendar.
The first event I added recurs three days a week, which is not an option.
The second event I added overlapped with the first in a way that glitched the renderer into showing it with the wrong time and duration.
😬 Gonna wait for V2.
In the near future, upgrades for Mathematica will be available only as part of a subscription to your license, as we will no longer offer upgrade-only purchases.
Today’s failed Mathematica experiment: ordering fruit by fiber per unit volume.
Interpreter["Food"]
? ✅EntityValue[#, "DietaryFiberContentPerServing"] / EntityValue[#, "DefaultServingSizeVolume"]
? ❌I read Stephen’s whole article on what’s new in Mathematica 14, as usual. Version 13 had so many new LLM features that I had to upgrade, but the only features in 14 that I think will affect my life are DateObject and Quantity getting 100–1,000x faster. Awesome, but I’ll probably skip this one.
The “Cognition” in Probabilistic Models of Cognition 📚 is the hardest part to get across when I recommend this book. Because the book isn’t “Bayes works, yawn.” It’s so much more interesting if you, like me, exist primarily in ML circles that focus on curve-fitting:
There are tiny, self-contained, editable code examples to demonstrate almost every point, and they run in-browser! I love this book.
I’ve been re-reading Probabilistic Models of Cognition 📚 for probably the entire decade that it’s existed. Every time, I notice bits that have been rewritten or added, and notice that I understand just a little bit more. And yesterday, I understood everything cover-to-cover for the first time! 🙌🏼