Here are two slices of the HSV cylinder for colours of the same value (~brightness).
The linear gradient in RGB space is on the left, the linear gradient in HSV space is on the right.
I think
@JoshWComeau
has figured it out! Here's the thought: gradients trace a line between two points in colour space, but there's more than one way to represent colours in a (typically 3D) space.
One way to pick out a colour is by specifying how much red, green, and blue it contains. These are RGB colour spaces.
Second pic is my phone screen under a microscope, which uses this idea of adding together R, G, and B
Another way to pick out a colour is by specifying its 'hue' (where on the colour wheel?), its saturation (how pastel vs colourful?), and its 'value' (how bright?). These are 'HSV' or 'HSB' colour spaces.
Josh's idea is that attractive-looking colour gradients are linear gradients in HSV-type colour spaces, and other gradients are unattractive because they trace a straight line through RGB-type spaces.
A puzzle:
Imagine you begin a journey in Seattle WA, facing exactly due east. Then start traveling forward, in a straight line along the Earth's surface.
This is because bad-looking gradients typically have a greyish bit between the two colours, while attractive gradients are colourful throughout (see above).
If you imagine the RGB cube, there is a channel of greyish colours running between the "all 0%" corner and the "all 100%" corner. Two colours with significantly different hues are going to have to pass through that channel.
In HSV space, two colours with significantly different hues avoid that grey zone by orbiting around it, remaining colourful.
Notice that just varying the hue to get between two colours of similar S and V means circling around vs cutting through the colour wheel.
Scott Aaronson: "I wrote a simple little program that would let people type either “f” or “d” and would predict which key they were going to push next. It’s actually very easy to write a program that will make the right prediction about 70% of the time."
Then, we found one student that the program predicted exactly 50% of the time. We asked him what his secret was and he responded that he 'just used his free will.'" Ha.
From 'Quantum Computing Since Democritus'
In Jan of this year,
@bryan_caplan
bet
@MatthewJBar
that no AI would reliably score an A on his economics midterm exams before 2029.
Three *months* later, GPT-4 scores an A.
In Jan of this year,
@bryan_caplan
bet
@MatthewJBar
that no AI would reliably score an A on his economics midterm exams.
Three *months* later, GPT-4 scores an A.
I like how photos of the moon passing in front of the Earth look totally fake. Hard to break our familiarity with atmospheric blur + eye-like focal lengths!
Yoshua Bengio admits to feeling "lost" over his life's work: "he would have prioritised safety over usefulness had he realised the pace at which it would evolve"
Life Universe
Explore the infinitely recursive universe of Game of Life! Works in real-time and is perfectly consistent, never fails to remember where you are and where you came from.
無限に再帰するライフゲームの宇宙を探索できる作品を作りました
#indiedev
A memorably insane detail from 'The Doomsday Machine' by the late Daniel Ellsberg:
In 1960, the US Air Force would sometimes task the RAND Corporation with assessing new technical proposals. One memo titled "Project Retro" fell to Ellsberg to assess.
In the original puzzle, you might have been imagining traveling a line of constant latitude — that is, *always* heading east.
But that is not a great circle, and so not a straight line: you'd need to be constantly turning left to maintain that path.
Continued: "I couldn’t even beat my own program, knowing exactly how it worked. I challenged people to try this and the program was getting between 70% and 80% prediction rates...
I think part of the trickiness here is that straight lines on the Earth are not straight lines on most 2D projections of the Earth (maps), and vice-versa.
Which is why the shortest flight path between two cities often looks unnecessarily curved on a map.
(To give an extreme example: imagine driving in a 10 meter radius around the North Pole. In order to always be traveling east — maintaining the same latitude — you'd need to be steering left the whole time)
The entire structure is aperiodic for space and time: it never perfectly resets, nor does it contain perfect copies of itself at any smaller scales. Which makes it especially tough to build a working demonstration without cheating. I'm in awe!
Should settling Mars be a current priority for longtermists? I think no, not by a long shot! (Let’s instead focus on preventing pandemics, decarbonising, making sure AGI doesn’t go terribly...)
Here’s why:
To state the obvious: it is really extraordinary how much more mainstream coverage of AI existential risk has recently become. All these stories in a ~2 month period!
Here's a better perspective: looking at Seattle centered on a globe, any straight line from Seattle will look like a straight line to you.
So with north pointing directly up, the path you take is a straight line directly to the right.
... the array of Atlas engines would be fired, as near simultaneously as possible, to stop the earth’s rotation momentarily.
The Soviet missiles, on their inertial path, would thus bypass or overfly their intended targets.
After pointing due east, and traveling continually forward in a straight line across the Earth's surface, the next country you hit after you leave North America is...
Australia.
I love data, but I also love space, so I'm excited to say we have a brand new collection of charts on space exploration & satellites on
@OurWorldInData
! 🧑🚀🚀🪐
12 charts on launches (number & cost), astronauts, exoplanets, asteroids, pollution, spending…
Another Morgenbesser anecdote:
Ordering dessert, the waitress tells him it's either apple pie or blueberry pie. Sidney orders the apple pie. Then the waitress returns to say she forgot that cherry pie is also an option.
"In that case", he responds, "I'll have the blueberry pie"
J. L. Austin: "Although a double negative in English implies a positive meaning, there is no language in which a double positive implies a negative."
Sidney Morgenbesser: "Yeah, yeah"
To be pedantic, by 'straight line' here I really mean 'geodesic' — the generalisation of 'straight line' for curved surfaces. but I think it's obvious enough.
If you're placed on a surface and continually walk "forward", by definition you will trace out a geodesic.
Very neat game to get intuitions for prompt injection: find ways to get LLM Gandalf to tell you the password for the level, except LLM Gandalf is instructed not to reveal the word 🧙
(I'm stuck on level 7)
Being a great circle, exactly half of the path you travel is in the Northern Hemisphere, and half in the Southern Hemisphere. It swoops under Africa to arrive at Australia somewhere near Perth.
Ok, what's going on here?
Forget about the Earth for a second, and just imagine you're standing on a big sphere.
Point in any direction, and begin walking forward along the surface of the sphere, without changing direction.
You'll loop round and return to where you started.
.
@OurWorldInData
has a couple new charts about space! 🚀
They show something striking: launch numbers completely plateaued for half a century, but have very recently *exploded*.
📈 Yearly:
📈 Cumulative:
A few months ago I was hanging out with friends in the park and (I can't remember how) we got to talking with an academic-looking older man sitting on a bench under a tree.
Before all the takes (limitations, worries, predictions etc) I think a really appropriate first reaction to advances like this is, and will continue to be —
"Wow"
Introducing Sora, our text-to-video model.
Sora can create videos of up to 60 seconds featuring highly detailed scenes, complex camera motion, and multiple characters with vibrant emotions.
Prompt: “Beautiful, snowy
Pangrams:
"The quick brown fox jumps over the lazy dog" → unnecesarily long; not even slightly ominous
"Sphinx of black quartz, judge my vow" → why did we go with the one about the fox when this exists?
There are some exceptions: any straight line on a 'gnomonic projection' (pictured) is an arc of a great circle (straight line) on a sphere.
But how natural to think a straight line on a more familiar map is a straight line on the surface of the Earth!
Notice, just intuitively, how you'll have walked the full circumference of the sphere: a 'great circle'.
In order to trace out a smaller circle than that, you'd need to be constantly veering left or right.
I think there's an 'ideas overhang' in EA.
The number of interesting / useful / important ideas has outgrown the number of ideas that have been communicated well, or at all.
Gold medal time in 100m final at the first modern Olympic games (1896): 12.0 seconds
Current record for fastest 100m by a pantomime horse (2009): 12.0 seconds
Progress in athletics!
New White House announcement on AI today.
Some promising signs: interest in the overlap with biosecurity, improving infosec for top labs, and some kind of initiative where labs are committing to model evaluations.
There's been talk recently about ulterior motives for expressing concern about advanced AI. But it's hard to read this as anything but a genuine, and emotionally difficult, and altogether inconvenient, change of mind
I got reminded of that encounter because my friend shared David's obituary from a couple days ago.
"[He] would have been shocked", said a friend of the family, "that he died quietly in his own bed".
RIP — what a life!
The 'centipede game' is one of the strangest things in game theory. Two players take turns to either pass on a growing pot to the other player, or take a slightly larger share of the existing pot for themselves. (1/11)
Ellsberg: "[it] proposed in some detail to assemble a huge rectangular array of one thousand first-stage Atlas engines—our largest rocket[s]—to be fastened securely to the earth in a horizontal position, facing in a direction opposite to the rotation of the earth...
Awesome artistic recreation of the Aztec city of Tenochtitlan.
I find it hard to imagine how present-day CDMX was raised from a 5000 square kilometre inland sea, since ~entirely drained. Some generation-spanning civil engineering!
a Portrait of Tenochtitlan, my 3D reconstruction of the capital of the Aztec Empire is released!
I've been looking forward to this for a long time, and I am really curious what all of you think.
Take a look:
#tenochtitlan
@bryan_caplan
@MatthewJBar
I forgot to mention the period of the bet, potentially making Bryan's bet look more vague than it was. The bet was that no AI would score an A by January 2029.
a major American industry that no one talks about is bundling alcoholism with other things so that people will pay without having to admit they're alcoholics
casinos, all inclusive resorts, vibe restaurants with crap food, guy who spent $32k on doordash - all alcoholism bundles
OpenAI announced their API for ChatGPT today.
1 million tokens processed for $2. It's 90% cheaper to run since December, and 10% the cost per token of their previous best GPT models.
Machine-generated text is already ~too cheap to meter...
... [it] envisioned that if our [radars] detected and reported on the huge viewing screens at NORAD a large flight of missile warheads coming across the North Pole from the Soviet Union...
Listening to a history audiobook while having breakfast. Learning that one of the great and unusual luxuries afforded to Charlemagne was to hear histories read aloud during his meals, from the finest tutors in the world. We've come far!
I was curious how GPT-4o would do on the ‘Reading the Mind in the Eyes Test’ (RMET), which gauges accuracy in recognising mental states from black-and-white photos of people's eyes.
It just got 26/36 when I tested it, which is roughly human-level.
Kevin Kelly (
@kevin2kelly
) is working on a book project trying to imagine feasible AI-enabled futures for humanity that are desirable.
It's sad and surprising to me how little serious attention this question gets!
A new kind of space race could be on, and things are moving fast. So shouldn’t we be paying more attention to the future of space governance? 🛰
Excited to share this problem profile I wrote for
@80000Hours
—
The scheme, which had already passed through multiple agencies without being discarded, was prompted by the worry that a surprise Soviet attack with ICBMs could incapacitate US land-based missiles before they could retaliate.
... Last year, the chip industry produced more transistors than the combined quantity of all goods produced by all other companies in all other industries in all human history."
Here's a quick observation about forecasting —
Sometimes the act of predicting the chance of something happening can affect the chance that it happens. You might think this most often makes it impossible to give an accurate guess. But that's not quite right! 🧵
Unclear if there are *ten* people in the world who (i) have experience with or a good practical understanding of US regulation / legislation and (ii) have thought significantly (2+ years) about risks from advanced AI