> Even in more literary sci-fi, for example Dune, the appeal is because it's basically just a medieval story transplanted into space. The focus is on politics and the human experience, not "what if storms lasted longer".
Yeah well if you dismiss more literary SF because it focuses on human experience rather than vibranium or whatever, it's not surprising that what remains has flat characters. Don't you see the circular logic?
> Regarding point 2: you can put your settings in a file `settings.typ` and import it from multiple files.
Let's say I have 3 flavors of settings and 10 different typ files - normally I'd just have 3 flavors of top.typ (top1.typ, top2.typ, top3.typ) with the correct settings for each flavor with settings proagated to all 10 files. Compiling top1/top2/top3 would then create flavor1.pdf, flavor2.pdf, and flavor3.pdf
Now how do I do it with settings1.typ, settings2.typ and settings3.typ? I have to go into the 10 different files and include the appropriate settings file! Or employ hacks like creating a common settings.typ using bash in the Makefile and including the common settings.typ in the 10 different files.
Edit: This is an actual use case - I'm helping with a resume, and have 3 different resume styles - a resume, a cv, and a timeline - and different files like education, work experience, honors, awards, publications, projects, etc and the level of detail, style, and what is included or not in each is controlled by which resume style is active. In latex I did this using \newcommand and the ifthenelse package.
In typst, I have had to resort to passing these global settings as arguments to functions spread across these different files, so each resume item (function) instantiated from the top file has a bunch of parameters like detail_level = 1, audited_courses = true, prefix_year = false, event_byline = true, include_url = true, etc., which make the functions unweildy.
> Just have a master settings.typ that you import in top1.typ, top2.typ and top3.typ?
Yes, but each included file (like education.typ, publications.typ, etc) should also get these settings propagated from top - which typst doesn't allow - the appropriate settings need to be included in each of these files.
> you can pass global settings at build time with `typst c --input name=value`
and hope that this will make settings available in education.typ. Because each .typ file is "pure" in the sense that it only knows the variables/functions that are defined in the file, or imported. This way you don't have a file magically affecting the bindings available in another file, which is nice.
It's true there are cases where you'd like something like the above. Currently you can do something like that using states and context (basically putting the "settings" into the document and retrieving that) but it's not so nice. In the future the plan is to make this nicer by allowing custom type definitions (and having show rules and set rules work with them as they work with built-in types).
The right way to approach this is to ask a question: What does 0.999... mean? What is the mathematical definition of this notation? It's not "what you get when you continue to infinity" (which is not clear). It's the value your are approaching as you continue to add digits.
When applying the correct definition for the notation (the limit of a sequence) there's no question of "do we ever get there?". The question is instead "can we get as close to the target as we want if we go far enough?". If the answer is yes, the notation can be used as another way to represent the target.
You can bypass the system prompt by using the API? I thought part of the "safety" of LLMs was implemented with the system prompt. Does that mean it's easier to get unsafe answers by using the API instead of the GUI?
> I can also say confidently that the #1 method to combat memory safety errors is array bounds checking. The #2 method is guaranteed initialization of variables. The #3 is stop doing pointer arithmetic (use arrays and ref's instead).
I think these are generally considered table stake in a modern programming language? That's why people are/were excited by the borrow checker, as data races are the next prominent source of memory corruption, and one that is especially annoying to debug.
>It explains why they singled out Reunion from France, it has a separate ccTLD
It also has a separate country abbreviation (RE). You know, like you'd see on an address. The thing that tells you where something, like a good imported in to the United States, is coming from.
This is is why it has a separate ccTLD by the way.
This blue sky thread is just an incredible example of motivated reasoning.
> It also has a separate country abbreviation (RE). You know, like you'd see on an address. The thing that tells you where something, like a good imported in to the United States, is coming from.
Yes, obviously, it's ISO 3166-1 but that's a batshit way of assigning tariffs. To the point I suspect it's a LLM.
Norfolk Island? The island with 3000 people, which in the context of international trade is a speck at the side of Australia. Or the uninhabited Heard Island and McDonald Islands with zero trade?
If Reunion and Norfolk Island are to be considered separately from their mainlands, where are the tariffs for Easter Island (Chile)? It has more people than Norfolk and probably more trade, it's 3700ish km from the administrative region it belongs, so it geographically distinct like Reunion.
Anyone (with a pulse) tasked with calculating the tariffs would see this and think "I have to remove these outliers". So the two options are:
A. Someone took the ISO 3166-1 codes and brainlessly calculated their batshit formula without noticing that HM doesn't produce anything. They did not instead do the more natural thing, go from highest imports to lowest which would've eliminated HM and most anomalies. They didn't even check their work.
B. They asked an LLM, which calculated this in the most naive way possible one-shot.
I dunno governor, this looks like vibecoded Excel spreadsheets.
>If Reunion and Norfolk Island are to be considered separately from their mainlands, where are the tariffs for Easter Island (Chile)?
Easter Island's mailing address says Chile. Norfolk Island mailing address does not include Australia, nor does Reunion include French.
>They asked an LLM, which calculated this in the most naive way possible one-shot.
Whats the connection between LLMs and domain name endings or whatever again? Like why does using them necessitate the use of LLMs?
Let's think through it from the ground up... How would you expect them to come up with a list of countries? Are they supposed to just get a group of people together and compile a list of countries they can think of by memory? Clearly they will refer to some standardized list.
Its worse than that. Its like saying you must have used chat gpt because you answered that 2+2= 4 and gasp so do the LLMs! Nevermind that its just the obvious answer to the question.
Lets see the prompt. The prompt further down in the thread that reproduces it was asking how to use tariffs to balance trade deficits with a 10% minimum. Is there any other answer then set the rate such that the deficit goes away or 10%, whichever is greater? No. That's just the answer to the question and is why ALL LLMs give the same answer.
I think that's a very unfair take. As a summary for non-experts I found it did a great job of explaining how by analyzing activated features in the model, you can get an idea of what it's doing to produce the answer. And also how by intervening to change these activations manually you can test hypotheses about causality.
It sounds like you don't like anthropomorphism. I can relate, but I don't get where Its a bit like there is the great and powerful man behind the curtain, lets trace the thought of this immaculate being you mere mortals is coming from. In most cases the anthropomorphisms are just the standard way to convey the idea briefly. Even then I liked how they sometimes used scare quotes as in it began "thinking" of potential on-topic words. There are some more debatable anthropomorphisms such as "in its head" where they use scare quotes systematically.
Also given that they took inspiration from neuroscience to develop a technique that appears successful in analyzing their model, I think they deserve some leeway on the anthropomorphism front. Or at least on the "biological metaphors" front which is maybe not really the same thing.
I used to think biological metaphors for LLMs were misleading, but I'm actually revising this opinion now. I mean I still think the past metaphors I've seen were misleading, but here, seeing the activation pathways they were able to identify, including the inhibitory circuits, and knowing a bit about similar structures in the brain I find the metaphor appropriate.
