rtpg

Yeah I did the French engineering school exams, I got a looooot of nasty grades in some of the test banks.

I was in the American school system up until HS and then moved to France. The _way_ stricter grading compared to what I had in the US was very hard to get used to.

One thing that I did witness while studying for the engineering school exams (which is a whole two year affair) was seeing a lot of people who were very used to being the top of their class and getting good grades flame out, because they had already been working super hard in high school to keep their grades up.

I think there's a good chunk of "straight A students" who might struggle with the hit to their ego in that environment. But it was helpful and humbling to me.

University objectives of making a diverse student body (an objective I basically agree with!) make this tricky and futzy (but not impossible IMO! You can always do something a bit hybrid), but I really feel like if some US university just decided to go to an entrance exam model they'd probably instantly have a cohort of students that would be _much_ easier to onboard into university materials.

Biggest problem is you'd need to make the entrance exams _actually tough_. Like "basically nobody who takes it gets an A". The objective is to rank students between each other, after all! It doesn't matter if the top scorer only has a 70% or something.

Every argument against standardized testing in this kind of thing seems to boil down to "test results are heavily correlated to socio-economic status". But... every other form of university entrance criteria process is even more correlated, right? All those extracurriculars aren't free after all.

That's the wrong read. It's more that it's easy to understand. "I will give you a 100 dollar bill, please give me a 10,000 yen bill."

I think the reason some people are complaining about the Suica stuff nowadays is that since... I think 2021 there's been stock shortages so getting a physical Suica involves going to specific places (I think at airports for example?) to get one. There's also now this "Welcome Suica" that you can get but it only works in certain areas of the country (so much for the IC standard...)

I believe you can't get a physical one from the ticket machines anymore unless you are getting a commuter pass, but I might be wrong.

I mean apart from the uploading of the backup xmpp already lets you do this right? That's my impression at least. Maybe I'm misunderstanding something.

(The premise in my understanding is that you own example.com and control its DNS records right? And not "well I hosted on foo.com as name@foo.com but it went down and now I want to move")

> Ideally, I'd like to have me@mydomain.com as my persistent XMPP ID that I can freely migrate across servers just by updating a DNS record without having to worry too much about who is actually hosting the server. The AT protocol does this right.

I suppose here you're saying "I should be able to do this without providers having to do work"? Like you show up with lxgr@foo.com to bar.com and create an account, and everything kinda fits into place (modulo you having to set up some DNS?)

It is already possible of course to have XMPP server at domain A handle things for domain B as well. But it is cool how ATProto seems to really bake in the concept of "all domains are welcome".

to anyone reading this who is thinking of setting up snikket: suffer through setting up prosody. It's a bit futzy but you won't find yourself saying "ah if only I had tried a bit harder" 10x a minute when trying to configure things later on.

I was generally "Matrix feels right" but I read through [0] and other discussions and I think that Matrix's E2E focus + way of duplicating everything just means that it suffers a lot of scalability issues that "plain old Jabber" won't have.

now Jabber has client issues, but stuff like Monal "basically work" for text + images + chat rooms. I'm excited for Fluux. You can of course set up one of the random web clients. Stuff like reactions "work", notifications is a solved problem.

But like... at a basic level, "I'm hosting a Jabber server for my small community" has a lot less headaches than with Matrix. Yes if your server goes down then you lose the chat. Oh no. Plenty survived off of IRC IMO.

For the HN crowd... I would say that there's still fun jank, but who doesn't like jankiness in their programs for fun? People should swarm to this opportunity to have some fun with busted tooling again!

[0] https://wiki.alopex.li/ADiveIntoOpenChat

Bit of a tangent but does anyone have good learning materials on doing things with Emacs Lisp beyond just staring at the source of packages?

I have the Mastering Emacs book and read through most of it, which gives me a good idea for some stuff... but I really would like to have good ideas for the right kind of design patterns for common things.

I think Qt is a bit different here because of Qt Quick etc but I've tried using "native" GUI frameworks a lot and ultimately the web DOM model is pretty nice for iterating easily.

Querying into the DOM and working off of that means you're not futzing about holding onto a bunch of component references for the one thing you might need later on.

Web layout is also pretty easy to just get right, native GUI frameworks have a different kind of layout model that is not as amenable to "arbitrary" data (at least at first blush).

the reactive programming model also works quite well. While there are native GUI frameworks that lean into reactive programming, they tend to be doing a bunch of weird stuff that cause other issues.

And at the end of the day, when you package something like a Qt app it still often ends up being quite chunky.

In the end tho... the simplest thing is if you do web stack you get a web UI _and_ your "native" GUI in one go. Build things once, not twice. Hard to argue against that when that's presented.

I'm having a bit of trouble reading this over understanding the case where I would rather the agent update itself, vs having the agent write out tools and then call those tools.

I rarely find myself thinking "ah I need the agent interface to change". And the vague generality of how agents work play into making it fairly easy to "just" have it rely on some external tooling to do anything special.

Maybe this is just a counterargument to the lisp philosophy as a whole but... well... I use Emacs for example and am fine with a model of "an agent can look at my Emacs config" rather than "my agent _is running my Emacs session_".

OTel has a nice property of working, which sometimes can't be said about vendor stuff.

I think the actual APIs kinda smell at the language level, and when Honeycomb decided to lean into otel and deprecate its Python libs I was super sad, cuz HC's libs were _way_ more usable IMO. Docs are also... painful. Real painful.

I wish that I could get a Python lib which is like "here this is Otel but the config phase isn't weird, and the API just looks a bit better". One of these days.

The biggest trouble I have with Otel recently is getting fixes patched upstream in contrib. Using contribs is super dangeerous, and I would basically recommend people write their own instrumentation and treat the contrib packages as just examples of how to do it

We use victoriametrics, but I believe that's just the collector side of it cuz we also query into it with grafana.

Datadog isn't just a collector, but the whole querying UI as well, right?

Let's say I type "SomeStructure" and haven't imported it, how does rust glancer _not_ end up poking at all the dependencies to figure out where it is, and to import it?

In some sense I feel like that example is the canonical "you need a full picture of the world" case, to the point that it's probably worth special casing _just_ that, and using local analysis for everything... but it seems hard to avoid that pain on any typo for example

Inversely... do you have a ballpark for what the size of the on-disk structures end up being? Not that it matters _that_ much but Rust projects already have a tendancy to be disk hogs during development

That's fair, at a previous place I worked at we had SFDC integration and there was definitely a lot of restrictions and pain points there. Especially if you're integrating, it's an uphill battle.

Don't you feel like it's still fairly far ahead of other systems, if you buy into the model?

I think there are other systems that kinda get there but SFDC is the only one of these kinds of systems that give you a lot of options on the backend and frontend side of things to just ship your own things and have them be integrated.

I think the real thing is that software for one _business_ is very much in demand (and a bunch of SaaS exists as a release valve for those use cases). And that stuff is hard to get working

I've been saying it to anyone who cares to listen: Salesforce is basically the biggest "Smalltalk" deployment out there. Super malleable, has a lot of ... decent rules for how to get software out into the world, every environment is basically its own universe at one point.

Someone should really attack SFDC from the "malleable Business OS" side of things. Unfrotunately everyone seems to attack it from the CRM side of things and lose the plot a bit.

Def agree that "have the thing I just came up with running in some environment that is sharable" is still a pretty gnarly problem, and LLM stuff doesn't... I don't think they've really made much progress on it.

At one point someone really needs to make an easy to manage box metaphor for all the stuff. It's not Docker. Google scripts is almost there but you really want to have the database be a thing you can drag and drop as well (SQLite people: get excited!)

Dropbox for business apps. If anyone runs with this at least send me an email thanking you for the metaphor.

Even just B2B vs B2C is a huge split. Plenty of specialized ERPs in the world that have relatively few "butts in seats" users but have a lot of data to work through.

Size of the data, read/write ratio, number of "requests"... lots of axes that change how much pain or not you're in by just pointing to pg

3 months since the last stable release feels ... alright. The constant promises for a new stable release feels a bit like a self-own.

Having said that I would have a bit more confidence in things if release just got back to a normal pace.

what is the visualisation strategy here? I find post-hoc grouping and plotting in pandas pretty satisfying.

do you have a snippet of what this looks like in R?

Pandas and Polars are both a place where I _really_ would love to have some sort of macro subsystem for Python. `pl.col(...)` is a neat trick for slicing, the pandas `df[df["foo"] == "bar"]` thing has always felt a bit of a mouthful (especially if you deign to use a longer name for your dataframe).

I appreciate Polars offering some alternative APIs for poking around in data, though. I feel like at some point someone will land on a _very_ nice to use API

The "obvious" one: supply chain chaos due to everyone and their dog spending as much as possible for AI boxen

Every investor being full on AI has lead to big corps massively increasing AI spend, instead of "just" hiring people (talk to people who graduated from school in the past year...)

I've seen some commentary about how SV turning from an "infinite cash reserves" place to a "let's massively load up on debt" place is unraveling a lot of the financial status quo (There's an only slightly curved line from this dynamic changing to East Asian currency pains, and then from there to impeding US debt crisis)

A big undercurrent of _entire demographics being driven insane_ by all the AI nonsense. People have always had problems with "fake" imagery/etc, but it used to take a bit of time. And it used to require a lot of time to convince yourself that you were a super genius by reading a bunch of Wikipedia and coming up with your own pet theory of the world.

There are rumors that Trump is getting shown AI vids on the state of the war.

And way low on the scale of things: every goddamn restaurant since 2024 using AI image gen, meaning you have no goddamn clue what the food you're ordering actually is. Or you do but you know that it's _not_ what is pictured. And the logos are all ugly.

I love it. I'm gonna indent anyways for clarity!

I'm... not a super fan of rhombus' emphasis on metaprogramming beyond supporting the language itself though.

I don't know how true that is nowadays, I think if you sign up to Enterprise Cloud you're in the same spot as everyone else

(maybe F500 people have their colocated box, but I have the impression that those are the "first class" seats (very expensive but _very few_) whereas enterprise cloud is the "business class" (loooooots of seats making a lot of money)

My feeling is that the entire category of research is bad, even if you can end up with good implementations.

Like 10 years ago I realized that facial recognition research was basically on the same level as weapons research, if not worse. So many nefarious usages.

Every researcher and product designer who leaned into this just has helped speedrun the world into a nasty place. But I guess I don’t have to type my passcode into my phone anymore!

does this mean that you can't do `eval`-like things in "production" systems?

like legit CPUs etc start failing. There's throttling but ultimately even that....

things might have gotten better though, I last had convos with people who did this maybe 6 years ago at this point.