unchar1

It's not just figuring out if a model is good at things, but is it good at the things I care about.

Using a targeted eval suite (like a test suite) tells us that.

Opening the site on my macbook shot the CPU usage to >50%.

Seems a bit ironic considering that it's supposed to be a specification on how a website should be.

What was the reasoning for this change?

Claude tends to disregard "NEVER do X" quite often, but funnily enough, if you tell it "Always ask me to confirm before going X", it never fails to ask you. And you can deny it every time

The good thing is that local models are catching up very fast.

Calculators are a particularly bad example for your case. There was absolutely hyperbole against calculators when they were introduced. [1]

With similar sentiment as well "They make us dumb" "Machines doing the thinking for us"

Cars were definitely seen as a fad. More accurately a worse version of a horse [2]

If you looked through your other examples, you'd see the same for those as well.

Some things start as fads, but only time will tell if they gain a place in society. Truthfully it's too early to tell for AI, but the arguments you're making, calling it a fad already don't stand up to reason

[1] https://www.newspapers.com/article/the-item/160697182/ [2]: https://www.saturdayeveningpost.com/2017/01/get-horse-americ...

This looks interesting.

One small feedback: It took me a while to figure out what it actually does. The homepage makes it look more like coolify or dokku.

We did an A/B test of an old SPA app, and a modern re-write using SSR and server-rendered pages.

By every performance metric, the new app was faster.

But we kept getting user feedback that the new site was "clunky" and "slow", even though we saw that the p90 was much lower on the new site. Most of our users asked us to enable a toggle to let them go back to the old "fast" site.

I'm not sure if this is a universal experience, but I think a lot of other sites that tried the CSR -> SSR move had similar experiences. It's just harder to talk about, since it goes against the usual narrative.

Would it make sense to have a per-country H1B cap?

You can write memory safe code in any language, but having a machine i.e. the Rust compiler check it for you is less error-prone than if a human does it.

Also if you look at the repo, only 3% of the codebase is in Assembly. IMO if >95% project is in Rust, you can definitely claim it's a Rust project.

Parts of it could probably already exists in the LLM's corpus, but being able to join them together to build new things is what makes "vibe coding" so useful.

Even though, from personal experience, at scale it still falls apart

Library and ecosystem support is a big thing. Rust being ML-ish, and having a large ecosystem is probably it's greatest strength

I see an option on the application form "I'm not UE citizen". Should that have read "EU Citizen" instead?

Also it's a little strange to ask for a passport photo, first time I've seen that.

The first thing that comes to mind is macros.

Your Node.js server then interprets that JavaScript and executes C++.

Umm...no? V8 specifically compiles it into machine code directly.

There used to be a pseudo-translation layer in the CrankShaftScript days, but that hasn't been true in almost a decade.

I can decompile the binary and still see the type information.

Also no. The de-compiler can _infer_ the types, much like how V8 tries to infer the type. But the actual type information is gone.

Even in languages like Java where most of the type information is preserved, some things are still lost (e.g. generic methods with dynamicinvoke)

To be clear, they are not "using" TypeScript, it's more accurate to say they are providing TypeScript bindings.

Interesting that you say it's more about providing "bindings", and not really "using". Much of the types in the svelte codebase are never exported outside of svelte, and they are only consumed internally.

The problem they were having was with transpilation, since the browser doesn't run JS.

From Rich Harris (months after svelte switched to JSDoc) [1]:

removing types from your own code is clownish, epically misguided behaviour, but whatever — to each their own

I would suggest going through the issues and PRs in the codebase to see how invested the Svelte team is in typescript.

TypeScript is not "real" static typing in the same sense as Go, Rust, C#.

That is true for Go, Rust, C#. But the same thing is also true for languages like C, and Generics in Java. I'm sure both of those languages have weak type systems, but are definitely statically typed.

I think the fact that type information is lost after being compiled isn't really a classifier for typed/non-typed. Ultimately it all comes down to machine code, and that certainly isn't typed either.

[1] https://x.com/Rich_Harris/status/1699490194565578882

Nue seems to be in an odd space. It's an untyped framework, built on top of "web standards".

But if you wanted web standards + web component, Lit already fills that space. If you want untyped JS or little/no JS at all, there's HTMX. Or if you're just tired of React, and want something faster + simpler, there's Svelte/Solid.

I'm not sure what problem Nue is uniquely solving.

From the link [3] you posted,

If you're rabidly anti-TypeScript and think that us doing this vindicates your position, I'm about to disappoint you.

Rich and the rest of the Svelte team are still using typscript, just through JSDoc + type definition files.

In contrast the Nue team seems to want to keep the view layer untyped.

From the parent comment

real static typing (like Rust or Go) shines in business logic where it counts

it seems they don't consider typescript to be "real" static typing.

USB microphones usually have audio converters built into them. So if you're going for a cheap wired headphones, it's usually better to go with one with a 3.5mm jack.

fish instructions:

set PATH $PATH ~/.npm-global/bin

Fish has some nice utilities for these type of set calls

set --append PATH ~/.npm-global/bin

Yaak is still open source under the MIT license. You can view, modify, and run the source code for both personal and commercial use. Licensing applies only to the prebuilt binaries.

That's a pretty interesting pricing strategy. I think REHL is the only other project I saw with this type of pricing.

Cool site!

Is there any way to see the top posts? It's really helpful when I'm trying to catch up on things. I know HN doesn't, but I can usually use hn.algolia.com to find the top posts.

According to their docs, SST uses Pulumi under the hood [1]

It's supposed to be a bit easier for developers to pick up, but you should be able to achieve the same thing with Pulumi AFAIU

[1] https://sst.dev/docs/#faq

I agree. But it's one reason why someone who is concerned about performance, may prefer classes by default.

I'm not sure if "C++ class" is the right term, but it will certainly compile into a class behind the scenes [1]

You can use d8 to check what the class structure ends up looking like [2]

---

[1] https://v8.dev/docs/hidden-classes

[2] https://v8.dev/docs/d8

Classes generally point you towards writing more performant code. Factory functions allow you to achieve the same performance, you just have to be a bit more careful not to cause a depot :)

For example,

1. field declarations [1] make sure that the fields are always initialized in the same order. That way most of your functions end up monomorphic, instead of being polymorphic [2]

2. Method declarations are also (almost) free, since you only pay for them once, during class initialization.

You also get a few other niceties such as private properties. You can emulate private properties with closures in factory functions but V8 has a hard time optimizing, unfortunately.

---

[1] https://developer.mozilla.org/en-US/docs/Web/JavaScript/Refe...

[2] https://www.builder.io/blog/monomorphic-javascript

I also feel this with X/Twitter, where everything feels like a promotion for a course or a book or a new product. Posting things because they are fun to write is a lost art.

The WSL section doesn't really make sense

But what you must know is that WSL has a lot of problems when compared to a full Linux OS. Some packages will not be up to date and sometimes will not be as functional as a Linux distro.

AFAIK it's just a regular linux distro (in a VM), so I'm not sure why it would have different packages?

That may be due to a combination of Malware detection + most unix programs not really written to take advantage of the features NTFS has to offer

This is a great talk on the topic https://youtu.be/qbKGw8MQ0i8?si=rh6WJ3DV0jDZLddn

It may be useful if you have data that originates from another source or if something outside of our system references your entity. In that case you need to keep some form of an externalRef, so it's usually easier to just use an id that you can control, for referencing both internally and externally.