The Aaronson Oracle

DivingForGold

About this page: The idea and the algorithm are Aaronson's, and the lineage is Hagelbarger's and Shannon's. This page is a teaching version of their work, and it is not affiliated with any of them. Prior web implementations exist and are credited in reference 6. Built by Spencer Stanton as plain HTML, CSS and JavaScript. No framework, no build step, no analytics, and nothing about you leaves your browser: your session is held in localStorage and nowhere else. (Scott's name is mentioned at the bottom)

pdpi

This is, effectively, the game that branch predictors in CPUs are playing with your code. Except that, in that version of the game, you win when the CPU guesses correctly.

Jaxan

I remember a rock paper scissors AI decades ago that did something similar.

mkmk

Such a cool concept and demonstration, but utterly exhausting to read the claudeslop.

cammikebrown

I stayed around 50% for hundreds of presses. Not sure what I did right/wrong.

kevinwang

Related (2023) https://news.ycombinator.com/item?id=38079512

PrimeOS32

37% predicted. Easy. Just press doubles trebles and quadruples more than feels natural.

cirrhosis

This submission was on the front page of HN yesterday (Show HN: How random are you?): https://news.ycombinator.com/item?id=50074909 The top comment: > This is a nice spin on the Aaronson oracle: > In a class I taught at Berkeley, I did an experiment where 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 And today we have a web app of this that seems fully generated by Claude.

ChaitanyaSai

I find these unreadable. How, if these are trained on writing, and then reinforced to appeal to people, do they end up with this writing style? Is AIs voting on AI writing how we arrived at this unbearable load-bearing moment?

js8

There used to be some studies on telepathy in the 60s and 70s that were wrong, because they used humans to pick objects rather than random generators. But humans do not make good random generators, and can be exploited even by other humans. For example, IIRC, humans tend to select some other option with slightly higher probability.

dmd

Am I just a weirdo or something? It was at chance for me after more than 100.

sigbottle

For some reason I thought this was going to be talking about Scott Aaronson's TCS contributions

nico

The implementation on that website is eerily good, I recommend everyone to check it out It also inspired me to build a little game based on it. Can AI can read you mind? https://jeffyclassify.com/blog/mind-reader/

stephencanon

I really don't think I was doing anything especially clever and it was running at 32% for me over 50 samples. That's much worse than chance.

hmokiguess

I managed to beat the game tapping to the beat of songs and alternating which song every time

hlieberman

An interesting oracle, but my god, the writing is so clearly AI, I just can't stand it. "runs down the center".

socializer

This is a lot of (LLM-generated) text, but there's a much simpler demo you can do for yourself: ask someone to type a "random" sequence of zeroes and ones, and then compare it to the output from /dev/urandom. One immediately obvious pattern you'll see is that humans not familiar with this test will have a strong aversion to generating runs of zeroes and ones. For example, you probably won't see 0-0-0-0 or 1-1-1-1 in the human-generated sample, even though it has a pretty high chance of appearing in truly random data. It just doesn't "feel" random, right? This has a practical application: if you're building a game with a component of randomness, people will often feel that the game is unfair or biased if they lose a number of times in a row. Tetris, famously, works around this by repeatedly taking a "bag" of seven possible tiles and shuffling it, rather than picking the next tile based on a dice roll. This way, the distribution over time is still random, but you never get more than two identical tetrominoes in a row.

gavinh

I am repulsed by the website's prose and aesthetics.

maxdo

36% not too great.

BoiledCabbage

What's also cool is if you take that n-gram frequency model, and scale it up from just 2 characters ('d' and 'f') to a full alphabet. And increase n towards a very large number then you can in theory do a good job of predicting sentences. Unfortunately you'll have a lot of your long sequences will only have 1 or 2 data points as n gets large, and you'll need an absurd amount of memory to store it all So if instead you replace it with a "compressed" model that doesn't actually store all the seen n-gram options, but instead predicts them, then you've more or less just re-invented LLMs.