hashmap

reducing consumption across the board isnt just unprofitable, it would mean everyone agreeing to overcome our biological gradients. i do not think it is possible for us to do, and evolution has not equipped us to do that as far as i can tell.

my semi-superstitious take is that the race to achieve ai is grounded in needing something that knows whats going on and is able to make decisions aligned to generational time horizons. whether that works out or not time will tell, but i get the sense a "good enough" ai is probably our best shot at saving us from ourselves. it's clear we can't do that on our own.

And yet I bet you leave your spam filter on instead of explaining to the spam what's wrong with it.

this landed precisely on like 3 weird bugs ive been hitting and solving in different stupid ways for dealing with things like sgd collapsing too many good answers into one bad answer, and gave me a real direction to try to fix the link missing in my own ml stuff. what timing. i have tried analytic solutions too and they're useful for like mapping prompts into memory geometry but from there ive ended up still having to use sgd. cause i think what happens is, sgd teaches the neural net both the geometry and how to navigate it. if you just teleport to the answer it doesnt learn how to walk.

> If Apple could make money from removable batteries, meaning there was a market for it and people wanted it over some other alternative, are you suggesting they are not smart enough to do the research and work necessary to accomplish that?

sort of missed the point. market dominance and lock-in means they already are the 800-lb gorilla, and removable batteries sit below where it'd move most people to switch

> The reality is people don't want it, at all.

lmao thats a good blither

https://www.androidauthority.com/removable-battery-poll-resu...

> Also, the lighting connector is better than USB in every way. Mandating an inferior technology is an odd choice.

right, except in the ways that matter and that people care about

that isnt how markets really work. you could say that if apple had two otherwise identical iphones except one has removable battery and one doesn't. but the enshittification cycle works via a ratcheting effect. once you achieve a certain level of dominance and lock-in, you can start getting away with all kinds of anti-consumer strategies to make more money and not get punished for it, and your competitors will follow suit. as long as you can ratchet above whatever detrimental thing you want to get away with is you'll probably be fine.

you can look at the lightning connector as an example. if you said "if people wanted usb connectivity they wouldn't buy iphones", nobody would take you seriously. and when apple was forced to switch, it absolutely didnt tank their sales because people just loved the lightning connector so much. the bad thing went away and it was great.

Yeah, softmax may have useful applications, but anytime you find yourself using the same hammer for everything that looks like a nail it's a bit of a red flag.

If you take instances of softmax that you find in training / inference and there turn out to be a few, and use other things like entmax or sparsemax you see across the board improvements. And like top1 often is just the best answer too, there's a reason why when you're doing tool calls temp=0 is the way to go. Like do you really want creative unicode tokens when writing bash commands. From what I can tell, most of the time softmax is the worst answer that works.

Much easier.

> easier than identifying text distilled from the words of almost everyone alive.

Well, there's more than that going on. AI generated text encodes a high-dimension navigational trajectory that guides the model through its geometry smoothly, like a trail of breadcrumbs. Human speech doesn't do that, it's jagged and jumps around the manifold, and probably doesn't even land on the manifold a lot of the time, and models can recognize the difference pretty quick.

working to directly advance a product used substantially to oppress people via surveillance or war crimes, when you have many other choices, is immoral. easy.

Oh, yeah $12k would not do it. For a UBI to work we would have to shift a significant portion of the concentrated wealth. I too was laid off long enough ago that by now I would be in a bad spot without help, also no mortgage or anything, and I don't travel or go out much. UBI of any degree would do something, but it would have to be much higher than $12k to tread water just due to rent alone. Aside from UBI we would likely need to decouple housing from profit, it has the same problem as healthcare. Demand for it is inelastic to a certain degree, everyone needs somewhere to live.

It does, and it should. With each iteration getting closer to the goalposts exposes the flaws in the goalposts, and then you try to make better goalposts. The problem people seem to have with the goalposts moving is they assume the goalpost makers either made good goalposts or thought they made good goalposts, but the actual process is "do the best we can at the moment and update when we get better information".

You never see it how. Like in terms of raw resources or political will?

if the original problem happened because it ignored something you told it, telling it to not ignore something is a category error. the determinism isn't added by the message you're sending it, it's in the enforcement mechanism. this should be set to keep firing until the condition is met. so, ralph pretty much.

to that end i would also word this entirely differently. i would have it be informative rather than taking that posture. "The test suite has not yet been run, and the turn cannot proceed until a test run has completed following source changes. This message will repeat as long as this condition remains unmet." something like that. and even that would still frame-lock it poorly. You want it to be navigating from the lens that it's on a team trying to make something good, and the only way for that to happen is to have receipts for tests after changes so we dont miss anything, so please try again.

> The language they do best with is the one with the largest corpus in the training set.

Not the case, strangely. They do best with Elixir. https://arxiv.org/pdf/2508.09101

if you think datacenters are a waste (they are), wait til you hear about department of war spending

well yes and no, i meant https://en.wikipedia.org/wiki/Orthogonal_Procrustes_problem which, yes, is named for that stretchster

The native way to skip all that is train a small thingy to map hidden state -> token/thingy you care about once per model family, or just do it once and procrustes over the state from the model you're using to whatever you made the map for.

at certain scales, reality has to win out over whatever ideal you have in your head about how things should be. facebook is massive, a lot of society is on it, and its a problem to make recourse invisible to people most affected by the thing stealing their attention.

yann lecun has been saying this for years. but its not a language really its an abstract geometric representation. so similar semantic meanings of sentences in different languages land in the same place in different models, just rotated.

adhd'er here too. maybe the practice is good, but it takes a lot of energy, which is finite. i find that leaning on my strengths gets me far, far better results than trying to get up to par with everyone else on things im bad at. if a tool just lets you get started, and you can breeze through getting started on things that you might otherwise just never even start, it seems like using the tool is the way to go.

ive been fighting the way my brain works my whole life, and only recently have i switched to trying to work with the way it wants to work. i get so many more things done that are important to me, and i get them done without the implicit "i need to flagellate myself with this thing i hate because there is something wrong with me" that comes with those fights.

and yeah, the ai's come with their own problems. but the trade is so exponentially in the direction of being worth it. even just the being a decent rubber duck aspect of them can keep me on a task when i would never otherwise hope to see it through.

if something like a popup appears that i didnt ask the page to do i snap close the page and never look at it again

oh neat ill check that one out. i dont get that much speedup from ssd/128gb unified vs vram if im doing like a predefined set of prompts, since i have it load it from disk anyway and im just doing one forward pass per prompt, and just like load part of it at a time. its a bit slower if im doing cpu inferencing but i only had to do that with one model so far.

but yeah on demand would be a lot of ssd churn so id just do it for testing or getting some hidden state vectors.

this is neat but to me seems like the circuitous path to just skipping autoregression, whereas the direct path is to just not do autoregression. get your answers from the one forward pass, and instead of backprop just do lookups and updates as the same operation.

im kind of wondering like what the ceiling would be on reasoning for something like the 1.5T models with the repeating technique, but they would take a long time to download. i think if you have them already it would take maybe an hour or so to check against a swath of prompts. whats the reasoningest open model at the moment?

my guess is that large models trained on large corpuses there is just some ceiling of "reasoning you can do" given the internal geometry implied by the training data, cause text is lossy and low-bandwidth anyway, and theres only really so much of it. past some point you just have to have models learning from real-world interactions and my guess is we're already kind of there.

dude thats sick! i tried it out and it works. theres a couple layers in there that are part of the voidy block that doesnt do much for the selected answer, so i narrowed it down to L48-53 where this model is mapping out its reasoning strategy, and repeated that twice, i got a big improvement over the original config (i chose some questions from atropos and claude code made some up so idk not like a real dataset).

so thats about %15 more compute per forward pass with 0 extra memory which is just nuts, so for a streaming or disk-based setup its just free better answers. def wasnt gonna think of this myself.

  config               layers   overall    delta          math     reasoning  word problems
  baseline                 80    0.5391  +0.0000        0.5850        0.6357        0.3500
  rys                      87    0.5452  +0.0061        0.6706        0.6000        0.2723
  cartographer_repeat_x2   92    0.7741  +0.2350        0.8455        0.8214        0.6000

looks like the model gets a second/third go at figuring out how to approach the problem and it gets better answers.

i tried a matrix of other configurations and stuff gets totally weird. like playing em through backwards in that block doesnt make much of a difference / order doesnt seem to matter (?!). doubling each layer got a benefit, but if i doubled the layers and doubled that block there was interference. doubling the block where the model is architecting/crystallizing its plans improves reasoning but at the cost of other stuff. other mixes of blocks showed some improvements for certain kinds of prompts but didnt stand out as much.

i literally suggested this metaphor earlier yesterday to someone trying to get agents to do stuff they wanted, that they had to set up their guardrails in a way that you can let the agents do what they're good at, and you'll get better results because you're not sitting there looking at them.

i think probably once you start seeing that the behavior falls right out of the geometry, you just start looking at stuff like that. still funny though.

these sort-of-lies might help:

think of the latent space inside the model like a topological map, and when you give it a prompt, you're dropping a ball at a certain point above the ground, and gravity pulls it along the surface until it settles.

caveat though, thats nice per-token, but the signal gets messed up by picking a token from a distribution, so each token you're regenerating and re-distorting the signal. leaning on language that places that ball deep in a region that you want to be makes it less likely that those distortions will kick it out of the basin or valley you may want to end up in.

if the response you get is 1000 tokens long, the initial trajectory needed to survive 1000 probabilistic filters to get there.

or maybe none of that is right lol but thinking that it is has worked for me, which has been good enough

Yeah, that's more or less what I'm getting at.

I can sort of see one angle for it, and the parent story kind of supports it. Bad software is a forcing function for good hardware - the worse that software has gotten in the past few decades the better hardware has had to get to support it. Such that if you actually tried like OP did, you can do some pretty crazy things on tiny hardware these days. Imagine what we could do on computers if they weren't so bottlenecked doing things they don't need to do.

if im not sitting on my right foot with left knee under my chin my thinking takes a hit, but i also have to constantly switch how im sitting so i dont get annoyed. its hard not to slouch/melt into whatever im sitting on and i think the only way to offset all that is the gym.

"DEI" is an inherent part of the system - being "against DEI" is simply a statement about what kind of "DEI" you actually want.