For the byte pair encoding (how those tokens get created) i think https://bpemb.h-its.org/ does a good job at giving an overview. after that i’d say self attention from 2017 is the seminal work that all of this is based on, and the most crucial to understand. https://jtlicardo.com/blog/self-attention-mechanism does a good job of explaining it. And https://jalammar.github.io/illustrated-transformer/ is probably the best explanation of a transformer architecture (llms) out there. Transformers are made up of a lot of self attention.

it does help if you know how matrix multiplications work, and how the backpropagation algorithm is used to train these things. i don’t know of a good easy explanation off the top of my head but https://xnought.github.io/backprop-explainer/ looks quite good.

and that’s kinda it, you just make the transformers bigger, with more weight, pluck on a lot of engineering around them, like being able to run code and making it run more efficientls, exploit thousands of poor workers to fine tune it better with human feedback, and repeat that every 6-12 month for ever so it can stay up to date.

Well each token has a vector. So ‘co’ might be [0.8,0.3,0.7] just instead of 3 numbers it’s like 100-1000 long. And each token has a different such vector. Initially, those are just randomly generated. But the training algorithm is allowed to slowly modify them during training, pulling them this way and that, whichever way yields better results during training. So while for us, ‘th’ and ‘the’ are obviously related, for a model no such relation is given. It just sees random vectors and the training reorganizes them tho slowly have some structure. So who’s to say if for the model ‘d’, ‘da’ and ‘co’ are in the same general area (similar vectors) whereas ‘de’ could be in the opposite direction. Here’s an example of what this actually looks like. Tokens can be quite long, depending how common they are, here it’s ones related to disease-y terms ending up close together, as similar things tend to cluster at this step. You might have an place where it’s just common town name suffixes clustered close to each other.

and all of this is just what gets input into the llm, essentially a preprocessing step. So imagine someone gave you a picture like the above, but instead of each dot having some label, it just had a unique color. And then they give you lists of different colored dots and ask you what color the next dot should be. You need to figure out the rules yourself, come up with more and more intricate rules that are correct the most. That’s kinda what an LLM does. To it, ‘da’ and ‘de’ could be identical dots in the same location or completely differents

plus of course that’s before the llm not actually knowing what a letter or a word or counting is. But it does know that 5.6.1.5.4.3 is most likely followed by 7.7.2.9.7(simplilied representation), which when translating back, that maps to ‘there are 3 r’s in strawberry’. it’s actually quite amazing that they can get it halfway right given how they work, just based on ‘learning’ how text structure works.

but so in this example, us state-y tokens are probably close together, ‘d’ is somewhere else, the relation between ‘d’ and different state-y tokens is not at all clear, plus other tokens making up the full state names could be who knows where. And tien there’s whatever the model does on top of that with the data.

for a human it’s easy, just split by letters and count. For an llm it’s trying to correlate lots of different and somewhat unrelated things to their ‘d-ness’, so to speak

Huh that actually does sound like a good use-case of LLMs. Making it easier to weed out cheaters.

They don’t look at it letter by letter but in tokens, which are automatically generated separately based on occurrence. So while ‘z’ could be it’s own token, ‘ne’ or even ‘the’ could be treated as a single token vector. of course, ‘e’ would still be a separate token when it occurs in isolation. You could even have ‘le’ and ‘let’ as separate tokens, afaik. And each token is just a vector of numbers, like 300 or 1000 numbers that represent that token in a vector space. So ‘de’ and ‘e’ could be completely different and dissimilar vectors.

so ‘delaware’ could look to an llm more like de-la-w-are or similar.

of course you could train it to figure out letter counts based on those tokens with a lot of training data, though that could lower performance on other tasks and counting letters just isn’t that important, i guess, compared to other stuff

I’d just offer refunds immediately, for everything

you can’t pay the bill? That’s ok, you get a refund.

you just wanted someone to shout at because Klarna sucks? Refund

You can’t log in? Believe it or not, also a refund.

one other use case where they’re helpful is ‘translation’. Like i have a docker compose file and want a helm chart/kubernetes yaml files for the same thing. It can get you like 80% there, and save you a lot of yaml typing.

Wont work well if it’s mo than like 5 services or if you wanted to translate a whole code base from one language to another. But converting one kind of file to another one with a different language or technology can work ok. Anything to write less yaml…

with less cold water coming down the US east coast, temperatures will rise. With more energy (heat) in that system, hurricanes will be more severe and frequent.

Of course there are. But I mean, women’s hormones do affect mood during the menstrual cycle (my wife certainly says she’s more iritable before her period), and afaik the hormone therapy is some of the same hormones, so it didn’t seem far fetched at all to me that it could play a role. hence me asking.

but could as well have been some deep seated anger at the world or similar, or something in between. Mostly I was just trying to think of reasons for why she might not be as bad as she was seeming, benefit of the doubt kind of thing.

I used to work with a trans woman who was a huge bitch, at least some of the time. Like actually shouting at coworkers for tiny mistakes, all-caps shouting in company chat at people trying to help with stuff, thinking she’s the smartest person in any room, that kind of stuff.

i’ve always wondered if she’s just a bitch or if at least some of it could be a side effect of hormone therapy? I mean, completely changing the hormones for your body must have some pretty dramatic effects in many areas and might take a long time until your body adjusts.

but a definitely won’t just ask ‘yo. Are you just a huge bitch or is it your medication’ in a corporate setting.

[edit] just for clarity, she started transitioning about 1 month after she joined that team and I left after about a year and a half, in part because of the mood on the team going to shit, among other reasons. But so I couldn’t compare to pre-hormone therapy or anything like that.

[edit2] thank you for all the replies, this was really enlightening and answered a lot of questions! Especially on a topic i feel is discussed less often, or at least I haven’t come across.

Just to be clear, I’m not defending the decision in any way. The issue here is with letting him go free, that’s disgusting and should not have happened.

The privacy part makes sense to me. If the purpose of prison/the justice system is resocialization, instead of punishment, then having your name all over the news just screws you over forever, even if in the eye of the law you did your time and don’t pose a risk anymore, regret your crime etc., and it increases the chance of them just going back to crime when they leave prison, due to a lack of options.

Because in Switzerland a criminal is still covered by the right to privacy, so unless the person in question is a public figure like a politician or the information was already leaked(among some other exceptions), it is not published. Same for the victim. Most you get is ‘John D. from Soandsotown, 34 years old (name known to the editorial department)’, if that.

Wouldn’t that just lead to splitting off of cheap companies, with pro-bono ceos that get paid more by the parent company through side channels? I don’t think there’s any fixing it with these kinds of laws, as they’ll just find loop holes to circumvent it.

maybe if companies were forced to be democratic so figurehead ceos could be ousted by the underpaid workers, but at that point it’s not capitalism, but socialism. and that’s how it usually goes imo, the workable solution to capitalism turns out to be not-capitalism

I’m not really sure I follow.

Just to be clear, I’m not justifying anything, and I’m not involved in those projects. But the examples I know concern LLMs customized/fine-tuned for clients for specific projects (so not used by others), and those clients asking to have confidence scores, people on our side saying that it’s possible but that it wouldn’t actually say anything about actual confidence/certainty, since the models don’t have any confidence metric beyond “how likely is the next token given these previous tokens” and the clients going “that’s fine, we want it anyways”.

And if you ask me, LLMs shouldn’t be used for any of the stuff it’s used for there. It just cracks me up when the solution to “the lying machine is lying to me” is to ask the lying machine how much it’s lying. And when you tell them “it’ll lie about that too” they go “yeah, ok, that’s fine”.

And making shit up is the whole functionality of LLMs, there’s nothing there other than that. It just can make shit up pretty well sometimes.

Get it to draw an ascii art nipple?

It’s always funny to me when people do add ‘confidence scores’ to LLMs, because it always amounts to just adding ‘say how confident you are with low, medium or high in your response’ to th prompt, and then you have made up confidences for made up replies. And you can tell clients that it’s just made up and not actual confidence, but they will insist that they need it anyways…

Und die anderen 27% sind Firmeninhaber oder hoffen, es zu werden?

there was a lot of controversy recently around this.

In short, the swiss military department decided on buying the F35 because they got a really good fixed price, instead of going with other contenders, like the French Rafale. Journalists asked “did you really? no other country has a fixed price and certainly not that cheap”, but they doubled down, even published a page on their homepage saying all the stuff the journalists are saying is wrong, gave press conferences confirming that it was a fixed price agreement etc. The US ambassador wrote some statement at some point almost confirming that this was true, without actually confirming.

Fast forward a bit, the leader of the military department resigned and a bit later it turned out that the F35s would be more expensive, switzerland did not in fact have a fixed price agreement and costs would be like 50% more expensive than originally planned, and this was all just a woopsie daisy little misunderstanding. After repeated inquiries and articles that said that this will be the case back when this was originally discussed.

Of course everyone involved already resigned before this came to light (one of them now works in the swiss embassy in the US) and well, the agreement was already reached, what can you do about it, tough luck…

https://www.republik.ch/2025/06/27/das-milliardenschwere-missverstaendnis is a really good write up (german)

The internet was a bubble (dotcom) that burst and still stayed around, both things can be true.

Wird normalerweise zu Ostasien gezählt, nicht Südostasien

I tend to agree with Schopenhauer(other than it sounding quite arrogant/condescending the way he puts it…):

The cheapest sort of pride is national pride; for if a man is proud of his own nation, it argues that he has no qualities of his own of which he can be proud; otherwise he would not have recourse to those which he shares with so many millions of his fellowmen. The man who is endowed with important personal qualities will be only too ready to see clearly in what respects his own nation falls short, since their failings will be constantly before his eyes. But every miserable fool who has nothing at all of which he can be proud adopts, as a last resource, pride in the nation to which he belongs; he is ready and glad to defend all its faults and follies tooth and nail, thus reimbursing himself for his own inferiority.