achatham

Lately I've been using RSVP (rapid serial visual presentation) speed reading, where you see a single word at a time, reading at 400-600wpm. You have to pay attention or you miss the words, so the distraction questions are mostly moot. On the downside, it's a lot harder to think about what you're reading, so I'm questioning whether it even counts as having read the book. I haven't loved any books I've read this way, which could be bad luck with the content but may be more about how I'm consuming it.

DragonBox Algebra has a similar concept but gamified. It has a cool progression: you get a new "power" each chapter, like factoring or negating. And it starts out with monster cards then gradually transitions to "x" and numerals. My kids have loved it, but it's hard to tell how much they learn when you can only make legal moves. When my older one learned the basic rules of algebraic manipulation and went back to it, it worked great at giving him a visual model to follow.

That said, the game doesn't let you do arbitrary equations, so you cap out when you beat the game. Excited to try this app out!

Yes, Gemini is very token efficient at video. It also has "lower resolution" options which can make it even cheaper if. With Gemini 3.1 flash lite an hour of video works out to $0.24 at the API rates.

What conclusion do you think I was drawing? I was just sharing an interesting quote relevant to the thread.

Oil was solely a lighting product at this point. The Teamsters were clearly not thinking 70 years into the future to stop automobiles. But I think the "monopoly" part of the quote is somewhat germain, even if it's just opinion of the author.

No idea about them trying to ban automobiles, but oil pipelines were invented to get around their friction. From _The Prize_, referencing the mid-1800s:

"From the first discoveries, teamsters, lashing their horses, had clogged the roads of the Oil Regions with their loads of barrels. They were more than just a physical bottleneck. Holding a monopoly position, they charged exorbitant rates; it cost more to move a barrel over a few miles of muddy road to a railway stop than to transport it by rail from western Pennsylvania all the way to New York. The teamsters’ stranglehold on transportation led to an ingenious effort to develop an alternative—transportation by pipeline."

The donation was also made through a donor advised fund (DAF), which means Musk didn't legally make the donation. I'm surprised he didn't lose on not having standing.

We do support YouTube Music and actually supported that before Spotify. But we only do ad-supported on Spotify and iHeartRadio (also paid Spotify).

Making my own epub reader with the kitchen sink of features I'd like. It's a speed-reading app first and foremost, using RSVP (rapid serial visual presentation, one word at a time). Also answers questions about the book with an LLM without spoilers, and can create illustrations. I've been reading _Mercy of the Gods_ lately, which has vivid descriptions of a bunch of alien races, but the pictures have done a great job supplementing my imagination. I've read more books in the past month than the last year, but we'll see if I keep it up.

https://github.com/achatham/epub_speedread

My personal project is illustrating arbitrary stories with consistent characters and settings. I've rewritten it at least 5 times, and Nano Banana has been a game-changer. My kids are willing to listen to much more sophisticated stories as long as it has pictures, so I've used it to illustrate text like Ender's Game. Unfortunately, it's getting harder to legally acquire books in a format you can feed to an LLM.

I first extract all the entities from the text, generate characters from an art style, and then start stitching them together into individual illustrations. It works much better with NB than anything else I tried before.

This was in place of reorganizing the codebase, which would have been the alternative. I've done such work in the past, and I've found it's a pretty rare skillet to optimize compilation speed. There's just a lot less input for the compiler to look at, as the useless transitive text is dropped.

And to be clear, it also speeds up the original compilation, but that's not as noticeable because when you're compiling zillions of separate compilation units with massive parallelism, you don't notice how long any given file takes to compile.

Clang modules. Sorry, didn't realize the distinction!

We use c++ modules at Waymo, inside the google monorepo. The Google toolchain team did all the hard work, but we applied it more aggressively than any team I know of. The results have been fantastic, with our largest compilation units getting a 30-40% speedup. It doesn't make a huge difference in a clean build, as that's massively distributed. But it makes an enormous difference for iterative compilation. It also has the benefit of avoiding recompilation entirely in some cases.

Every once in a while something breaks, usually around exotic use of templates. But on the whole we love it, and we'd have to do so much ongoing refactoring to keep things workable without them.

Update: I now recall those numbers are from a partial experiment, and the full deployment was even faster, but I can't recall the exact number. Maybe a 2/3 speedup?

At Waymo we use c++ modules via clang and got the demanded 5x speedup.

As the article mentions, you need a close relationship between the compiler and build system, which Google already has. The google build tooling team got modules to mostly work but only turned them on in limited situations. But we but the bullet and turned them on everywhere, which has sped up compilation of individual files by more than 5x (I forget the exact number).

The remaining problem is that sometimes we get weird compilation errors and have to disable modules for that compilation unit. It's always around templates, and Eigen has been gnarly to get working.

> “While TBD Labs is still relatively new, we believe it has the greatest compute-per-researcher in the industry, and that will only increase,” Meta said.

Well, two ways to make that true!

And imagine a hypothetical project that's 75% excavation. It'd never be built today, but if excavation gets cheaper the project could be feasible.

And then an explosion of underground bunkers and volcano lairs.

As a counter to your one example:

I've worked on autonomous vehicles for 16 years and my largest philanthropic effort is improving public transit. The common theme is being really interested in transportation and wanting it to work well for people.

Cruise was also the top funder of one San Francisco's recent MUNI funding ballot propositions (which just barely failed). You can certainly have a cynical take on that, but they still did it.

The paper under discussion only considers human accidents in similar environments to where Waymo operates. So it's only making a claim about like-for-like driving.

You could still say you care about snow driving and want to see that comparison, but it doesn't mean the claims in this paper are wrong.

What you're describing is L4. L4 is fully autonomous but with limitations on where/when it can operate. Level 5 is that but without restrictions.

Level 2 and 3 are the mostly-automated version, and they differ in how much notice they're supposed to provide and how much attention they require.

I think you may mean Level 4. The difference between 4 and 5 is that 5 doesn't have any territory/environmental constraints, but you said you don't mind those.

Your link literally says the opposite. Sales dropped from 2023 to 2024.

But the Bakersfield/Merced HSR line is an actual project, estimated at over $30B (it makes no sense independently but is part of LA-to-SF).

You seem to be comparing an actual, private tech project to what you wished public transit looked like, not what it actually looks like.

Public transit is awesome, but construction costs in the anglophone world are bananas (https://transitcosts.com/).

We started January 2009.

FYI that article is an interview with Jo Boaler, the main architect of removing middle school Algebra in SF and California (attempted). Her views are pretty controversial among those promoting math excellence.

That's not to say there isn't anything worthwhile in the article, but figured people would want that context.

What's your evidence that it would crash without the remote operation, or that it's more than 1.0 people?

I built my kids "the box", which had a similar concept. It was an upholstered box with a frosted acrylic top and a camera mounted inside facing up. You would put cards on top of the box and it would recognize the symbols and take an action, like play a song or read a word. I used large printouts of Data Matrix, as I found that faster and most accurate to decode. The front (up-facing) side would have a colorful design.

In the most advanced version I had it read letters to read a word aloud (eg, a spelling tool). All so I could reenact the scene from Sneakers where they rearrange Scrabble tiles from SETEC ASTRONOMY to TOO MANY SECRETS :)

I've been doing this for the past 2 years with my two older kids (now 5.5 and 3.5) based on the same reddit post this article is about. I've found it works great, though I have high-focus kids. I call it "word cards".

For the older kid, he learned the letters at preschool and then we did the Bob Books together. That worked well for phonics, but we hit a wall with sight words that require memorization. That's where Anki saved the day. We'd do 5 minutes of word cards every morning, and he'd get a token on a reward tracker, which eventually added up to a toy. Worked great for both of us. He's now a fluent reader, and I add unusual words we encounter to the deck, though I've stopped pushing so hard on reading and we just do "fact cards".

For my middle child, I started with Teach Your Child to Read in 100 Easy Lessons (TYCTR100), in plain book form. That got her sounding out simple words at 2.5, but it's a really boring process, and I couldn't get her to do 20 minute lessons on a regular basis and didn't want to force it on her. But then I bought a set of TYCTR100 flashcards on Etsy and turned them into Anki cards, and that's worked great. Again, 5 minutes most mornings, though at her age it works better with a short term reward (eg, interleaving with Daniel Tiger).

Apparently my kids said they planned to do "word cards" with their kids when I wasn't around.

Anki is great at what it does, but the main enabler is finding a process your kids will pay attention to. I probably lucked out there.

The first big investments in petroleum were for lighting via relatively clean-burning kerosene. Petroleum as a fuel came later, when they had a lot kerosene byproduct lying around. There's a obvious problem with solar as a lighting solution: you also need good batteries.

It is in California, but it hasn't been enforced yet: https://www.autobodynews.com/news/tesla-not-allowed-to-call-...

But isn't that the "relative" point? A house in Santa Clara doesn't cost $1.4M because it's 3x better than a house in another state, or costs 3x as much to build. It's because there are a lot of other high-salary people in the area who need a house and not a lot of houses being built.

What's the trigger for all the diffusion video projects this week? Are they all at the same conference, or did one trigger everyone to rush to publish? I'm grateful but curious. Animating still pictures from my kids' books is my goal and it seems close to plausible just from this week's advances.