Architeuthis @awful.systems It's not always easy to distinguish between existentialism and a bad mood.
23-2
Leaving something to run for 20-30 minutes expecting nothing and actually getting a valid and correct result: new positive feeling unlocked.
Now to find out how I was ideally supposed to solve it.
If nothing else, you've definitely stopped me forever from thinking of jq as sql for json. Depending on how much I hate myself by next year I think I might give kusto a shot for AOC '25
22-2 commentary
I got a different solution than the one given on the site for the example data, the sequence starting with 2 did not yield the expected solution pattern at all, and the one I actually got gave more bananas anyway.
The algorithm gave the correct result for the actual puzzle data though, so I'm leaving it well alone.
Also the problem had a strong map/reduce vibe so I started out with the sequence generation and subsequent transformations parallelized already from pt1, but ultimately it wasn't that intensive a problem.
Toddler's sick (but getting better!) so I've been falling behind, oh well. Doubt I'll be doing 24 & 25 on their release days either as the off-days and festivities start kicking in.
I mean, you could have answered by naming one fabled new ability LLM's suddenly 'gained' instead of being a smarmy tadpole, but you didn't.
What new AI abilities, LLMs aren't pokemon.
Slate Scott just wrote about a billion words of extra rigorous prompt-anthropomorphizing fanfiction on the subject of the paper, he called the article When Claude Fights Back.
Can't help but wonder if he's just a critihype enabling useful idiot who refuses to know better or if he's being purposefully dishonest to proselytize people into his brand of AI doomerism and EA, or if the difference is meaningful.
edit: The claude syllogistic scratchpad also makes an appearance, it's that thing where we pretend that they have a module that gives you access to the LLM's inner monologue complete with privacy settings, instead of just recording the result of someone prompting a variation of "So what were you thinking when you wrote so and so, remember no one can read what you reply here". Que a bunch of people in the comments moving straight into wondering if Claude has qualia.
Rationalist debatelord org Rootclaim, who in early 2024 lost a $100K bet by failing to defend covid lab leak theory against a random ACX commenter, will now debate millionaire covid vaccine truther Steve Kirsch on whether covid vaccines killed more people than they saved, the loser gives up $1M.
One would assume this to be a slam dunk, but then again one would assume the people who founded an entire organization about establishing ground truths via rationalist debate would actually be good at rationally debating.
It's useful insofar as you can accommodate its fundamental flaw of randomly making stuff the fuck up, say by having a qualified expert constantly combing its output instead of doing original work, and don't mind putting your name on low quality derivative slop in the first place.
16 commentary
DFS (it's all dfs all the time now, this is my life now, thanks AOC) pruned by unless-I-ever-passed-through-here-with-a-smaller-score-before worked well enough for Pt1. In Pt2 in order to get all the paths I only had to loosen the filter by a) not pruning for equal scores and b) only prune if the direction also matched.
Pt2 was easier for me because while at first it took me a bit to land on lifting stuff from Djikstra's algo to solve the challenge maze before the sun turns supernova, as I tend to store the paths for debugging anyway it was trivial to group them by score and count by distinct tiles.
And all that stuff just turned out to be true
Literally what stuff, that AI would get somewhat better as technology progresses?
I seem to remember Yud specifically wasn't that impressed with machine learning and thought so-called AGI would come about through ELIZA type AIs.
In every RAG guide I've seen, the suggested system prompts always tended to include some more dignified variation of "Please for the love of god only and exclusively use the contents of the retrieved text to answer the user's question, I am literally on my knees begging you."
Also, if reddit is any indication, a lot of people actually think that's all it takes and that the hallucination stuff is just people using LLMs wrong. I mean, it would be insane to pour so much money into something so obviously fundamentally flawed, right?
Pt2 commentary
I randomly got it by sorting for the most robots in the bottom left quadrant while looking for robot concentrations, it was number 13. Despite being in the centre of the grid it didn't show up when sorting for most robots in the middle 30% columns of the screen, which is kind of wicked, in the traditional sense.
The first things I tried was looking for horizontal symmetry (find a grid where all the lines have the same number of robots on the left and on the right of the middle axis, there is none, and the tree is about a third to a quarted of the matrix on each side) and looking for grids where the number of robots increased towards the bottom of the image (didn't work, because turns out tree is in the middle of the screen).
I thinks I was on the right track with looking for concentrations of robots, wish I'd thought about ranking the matrices according to the amount of robots lined up without gaps. Don't know about minimizing the safety score, sorting according to that didn't show the tree anywhere near the first tens.
Realizing that the patterns start recycling at ~10.000 iterations simplified things considerably.
The tree on the terminal output
(This is three matrices separated by rows of underscores)
13 commentary
Solved p1 by graph search before looking a bit closer on the examples and going, oh...
In pt2 I had some floating point weirdness when solving for keypress count, I was checking if the key presses where integers (can't press button A five and half times after all) by checking if A = floor(A) and sometimes A would drop to the number below when floored, i.e. it was in reality (A-1).999999999999999999999999999999999999999999999. Whatever, I rounded it away but I did spend a stupid amount of time on it because it didn't happen in the example set.
If you never come up with a marketable product you can remain a startup indefinitely.
Getting Trump reelected should count as 'billionaire philanthropy'.
thinkers like computer scientist Eliezer Yudkowsky
That's gotta sting a bit.
11 discussion with spoliers
Well my pt1 solution would require something like at least 1.5 petabytes RAM to hold the fully expanded array, so it was back to the drawing board for pt2 😁
Luckily I noticed the numbers produced in every iteration were incredibly repetitive, so I assigned a separate accumulator to each one, and every iteration I only kept the unique numbers and updated the corresponding accumulators with how many times they had appeared, and finally I summed the accumulators.
The most unique numbers in one iteration were 3777, the 75 step execution was basically instant.
edit: other unhinged attempts included building a cache with how many pebbles resulted from a number after x steps that I would start using after reaching the halfway point, so every time I found a cached number I would replace that branch with the final count according to the remaining steps, but I couldn't think of a way to actually track how many pebbles result downstream from a specific pebble, but at least it got me thinking about tracking something along each pebble.
11 code
// F# as usual
// fst and snd are tuple deconstruction helpers
[<TailCall>]
let rec blink (idx:int) (maxIdx:int) (pebbles : (int64*int64) list) =
if idx = maxIdx
then pebbles |> List.sumBy snd
else
pebbles
// Expand array
|> List.collect (fun (pebbleId, pebbleCount) ->
let fpb = float pebbleId
let digitCount = Math.Ceiling(Math.Log(fpb + 1.0,10))
match pebbleId with
| 0L -> [ 1L, pebbleCount ]
| x when digitCount % 2.0 = 0.0 ->
let factor = Math.Pow(10,digitCount/2.0)
let right = fpb % factor
let left = (fpb - right) / factor
[int64 left, pebbleCount; int64 right,pebbleCount]
| x -> [ x * 2024L, pebbleCount ])
// Compress array
|> List.groupBy fst
|> List.map (fun (pebbleId, pebbleGroup) -> pebbleId, pebbleGroup |> List.sumBy snd)
|> blink (idx+1) maxIdx
"./input.example"
|> Common.parse
|> List.map (fun pebble -> pebble,1L)
|> blink 0 25
|> Global.shouldBe 55312L
"./input.actual"
|> Common.parse
|> List.map (fun pebble -> pebble,1L)
|> blink 0 75
|> printfn "Pebble count after 75 blinks is %d"
promise me you’ll remember me
I'm partial to Avenge me! as last words myself.
puzzles unlock at 2PM
Almost exactly at the start of office hours for me.
Advent of Code 2024 - Historian goes looking for history in all the wrong places
copy pasting the rules from last year's thread:
>Rules: no spoilers.
>The other rules are made up aswe go along.
>Share code by link to a forge, home page, pastebin (Eric Wastl has one here) or code section in a comment.
New article from reflective altruism guy starring Scott Alexander and the Biodiversity Brigade
This post discusses the influence of human biodiversity theory on Astral Codex Ten and other work by Scott Alexander.
Would've been way better if the author didn't feel the need to occasionally hand it to siskind for what amounts to keeping the mask on, even while he notes several instances where scotty openly discusses how maintaining a respectable facade is integral to his agenda of infecting polite society with neoreactionary fuckery.
It can't be that the bullshit machine doesn't know 2023 from 2024, you must be organizing your data wrong (wsj)
>AI Work Assistants Need a Lot of Handholding
> Getting full value out of AI workplace assistants is turning out to require a heavy lift from enterprises. ‘It has been more work than anticipated,’ says one CIO.
aka we are currently in the process of realizing we are paying for the privilege of being the first to test an incomplete product.
>Mandell said if she asks a question related to 2024 data, the AI tool might deliver an answer based on 2023 data. At Cargill, an AI tool failed to correctly answer a straightforward question about who is on the company’s executive team, the agricultural giant said. At Eli Lilly, a tool gave incorrect answers to questions about expense policies, said Diogo Rau, the pharmaceutical firm’s chief information and digital officer.
I mean, imagine all the non-obvious stuff it must be getting wrong at the same time.
> He said the company is regularly updating and refining its data to ensure accurate results from AI tools accessing it. That process includes the organization’s data engineers validating and cleaning up incoming data, and curating it into a “golden record,” with no contradictory or duplicate information.
Please stop feeding the thing too much information, you're making it confused.
> Some of the challenges with Copilot are related to the complicated art of prompting, Spataro said. Users might not understand how much context they actually need to give Copilot to get the right answer, he said, but he added that Copilot itself could also get better at asking for more context when it needs it.
Yeah, exactly like all the tech demos showed -- wait a minute!
> [Google Cloud Chief Evangelist Richard Seroter said] “If you don’t have your data house in order, AI is going to be less valuable than it would be if it was,” he said. “You can’t just buy six units of AI and then magically change your business.”
Nevermind that that's exactly how we've been marketing it.
Oh well, I guess you'll just have to wait for chatgpt-6.66 that will surely fix everything, while voiced by charlize theron's non-union equivalent.
Generating (often non-con) porn is the new crypto mining
And now I feel like I need to take a shower. Ugh.
An AI company has been generating porn with gamers' idle GPU time in exchange for Fortnite skins and Roblox gift cards
> "some workloads may generate images, text or video of a mature nature", and that any adult content generated is wiped from a users system as soon as the workload is completed.
> However, one of Salad's clients is CivitAi, a platform for sharing AI generated images which has previously been investigated by 404 media. It found that the service hosts image generating AI models of specific people, whose image can then be combined with pornographic AI models to generate non-consensual sexual images.
Investigation link: https://www.404media.co/inside-the-ai-porn-marketplace-where-everything-and-everyone-is-for-sale/
SBF's effective altruism and rationalism considered an aggravating circumstance in sentencing
Sam Bankman-Fried maintains that his crimes were victimless and resulted in zero losses, and therefore warrant only six years of imprisonment. Prosecutors argue that 40–50 years are justified.
For thursday's sentencing the us government indicated they would be happy with a 40-50 prison sentence, and in the list of reasons they cite there's this gem:
> 4. Bankman-Fried's effective altruism and own statements about risk suggest he would be likely to commit another fraud if he determined it had high enough "expected value". They point to Caroline Ellison's testimony in which she said that Bankman-Fried had expressed to her that he would "be happy to flip a coin, if it came up tails and the world was destroyed, as long as if it came up heads the world would be like more than twice as good". They also point to Bankman-Fried's "own 'calculations'" described in his sentencing memo, in which he says his life now has negative expected value. "Such a calculus will inevitably lead him to trying again," they write.
Turns out making it a point of pride that you have the morality of an anime villain does not endear you to prosecutors, who knew.
Bonus: SBF's lawyers' list of assertions for asking for a shorter sentence includes this hilarious bit reasoning:
> They argue that Bankman-Fried would not reoffend, for reasons including that "he would sooner suffer than bring disrepute to any philanthropic movement."
Rationalist org bets random substack poster $100K that he can't disprove their covid lab leak hypothesis, you'll never guess what happens next
rootclaim appears to be yet another group of people who, having stumbled upon the idea of the Bayes rule as a good enough alternative to critical thinking, decided to try their luck in becoming a Serious and Important Arbiter of Truth in a Post-Mainstream-Journalism World.
This includes a randiesque challenge that they'll take a $100K bet that you can't prove them wrong on a select group of topics they've done deep dives on, like if the 2020 election was stolen (91% nay) or if covid was man-made and leaked from a lab (89% yay).
Also their methodology yields results like 95% certainty on Usain Bolt never having used PEDs, so it's not entirely surprising that the first person to take their challenge appears to have wiped the floor with them.
Don't worry though, they have taken the results of the debate to heart and according to their postmortem blogpost they learned many important lessons, like how they need to (checks notes) gameplan against the rules of the debate better? What a way to spend 100K... Maybe once you've reached a conclusion using the Sacred Method changing your mind becomes difficult.
I've included the novel-length judges opinions in the links below, where a cursory look indicates they are notably less charitable towards rootclaim's views than their postmortem indicates, pointing at stuff like logical inconsistencies and the inclusion of data that on closer look appear basically irrelevant to the thing they are trying to model probabilities for.
There's also like 18 hours of video of the debate if anyone wants to really get into it, but I'll tap out here.
quantian's short writeup on the birdsite, will post screens in comments
pdf of judge's opinion that isn't quite book length, 27 pages, judge is a microbiologist and immunologist PhD
pdf of other judge's opinion that's 87 pages, judge is an applied mathematician PhD with a background in mathematical virology -- despite the length this is better organized and generally way more readable, if you can spare the time.
rootclaim's post mortem blogpost, includes more links to debate material and judge's opinions.
edit: added additional details to the pdf descriptions.
Hi, I'm Scott Alexander and I will now explain why every disease is in fact just poor genetics by using play-doh statistics to sorta refute a super specific point about schizophrenia heritability.
edited to add tl;dr: Siskind seems ticked off because recent papers on the genetics of schizophrenia are increasingly pointing out that at current miniscule levels of prevalence, even with the commonly accepted 80% heritability, actually developing the disorder is all but impossible unless at least some of the environmental factors are also in play. This is understandably very worrisome, since it indicates that even high heritability issues might be solvable without immediately employing eugenics.
Also notable because I don't think it's very often that eugenics grievances breach the surface in such an obvious way in a public siskind post, including the claim that the whole thing is just HBD denialists spreading FUD:
> People really hate the finding that most diseases are substantially (often primarily) genetic. There’s a whole toolbox that people in denial about this use to sow doubt. Usually it involves misunderstanding polygenicity/omnigenicity, or confusing GWAS’ current inability to detect a gene with the gene not existing. I hope most people are already wise to these tactics.
Reply guy EY attempts incredibly convoluted offer to meet him half-way by implying AI body pillows are a vanguard threat that will lead to human extinction...
... while at the same time not really worth worrying about so we should be concentrating on unnamed alleged mid term risks.
EY tweets are probably the lowest effort sneerclub content possible but the birdsite threw this to my face this morning so it's only fair you suffer too. Transcript follows:
Andrew Ng wrote: >In AI, the ratio of attention on hypothetical, future, forms of harm to actual, current, realized forms of harm seems out of whack. > >Many of the hypothetical forms of harm, like AI "taking over", are based on highly questionable hypotheses about what technology that does not currently exist might do. > >Every field should examine both future and current problems. But is there any other engineering discipline where this much attention is on hypothetical problems rather than actual problems?
EY replied: >I think when the near-term harm is massive numbers of young men and women dropping out of the human dating market, and the mid-term harm is the utter extermination of humanity, it makes sense to focus on policies motivated by preventing mid-term harm, if there's even a trade-off.
Turns out Altman is a lab-leak covid truther, calls virus 'synthetic' according to Spectator piece on AI risk.
> Sam Altman, the recently fired (and rehired) chief executive of Open AI, was asked earlier this year by his fellow tech billionaire Patrick Collison what he thought of the risks of synthetic biology. ‘I would like to not have another synthetic pathogen cause a global pandemic. I think we can all agree that wasn’t a great experience,’ he replied. ‘Wasn’t that bad compared to what it could have been, but I’m surprised there has not been more global coordination and I think we should have more of that.’
Rationalist literary criticism by SBF, found on the birdsite
original is here, but you aren't missing any context, that's the twit.
> I could go on and on about the failings of Shakespear... but really I shouldn't need to: the Bayesian priors are pretty damning. About half the people born since 1600 have been born in the past 100 years, but it gets much worse that that. When Shakespear wrote almost all Europeans were busy farming, and very few people attended university; few people were even literate -- probably as low as ten million people. By contrast there are now upwards of a billion literate people in the Western sphere. What are the odds that the greatest writer would have been born in 1564? The Bayesian priors aren't very favorable.
edited to add this seems to be an excerpt from the fawning book the big short/moneyball guy wrote about him that was recently released.
Quality sneer found on the birdsite
Transcription:
Thinking about that guy who wants a global suprasovereign execution squad with authority to disable the math of encryption and bunker buster my gaming computer if they detect it has too many transistors because BonziBuddy might get smart enough to order custom RNA viruses online.