sbrother

This is very cool! Thanks for sharing.

That sounds great. How well does the base OS + "pet" containers work with all the crazy dependencies you need to do modern ML work, e.g. some exact combination of nvidia drivers + CUDA + torch + other random stuff? That's the pain point I'd be motivated enough to solve that I'd switch distros.

Some forms of kernel anticheat make dual booting harder, too. I can’t play valorant since that version of Vanguard requires secure boot, which doesn’t seem to work with my dual boot setup unless I invest more time fiddling than I care to. Easier just not to play that game.

Wait, is that a product that one can buy? It sounds like it would solve most of my issues with Windows.

That's the workaround I've used. I still want to be able to just press command F and see "error" highlighted throughout the scrollback.

Is there a blog post anywhere that explains why it has taken so long to get search working? Ghostty is such a nice app but that's such a fundamental feature that there has to be a good reason... It's the only thing keeping me on iTerm2.

Thank you; I feel similarly and wish I could go back as an adult to take even more advantage of all the incredible opportunities. Having four years to dedicate to learning new things in depth -- with zero pressure to take shortcuts so that the lessons become economically useful sooner -- was more of a privilege than I understood at the time.

And then of course, the people I met there have shaped my life and career in wonderful ways ever since. The sheer level of diversity among students and faculty is unlike anything I've experienced elsewhere. Many of them are still my lifelong friends (or in one case my wife :)) and others have opened professional doors to me 15 years later and counting.

But also, I went to a very well known and respected university with sufficient endowment and financial aid that it shouldn't be functioning as a "toll gate" regardless. I know things are not this rosy at a lot of universities.

How good is the interop story with Elixir/Erlang currently? Can I include a few gleam modules in my Elixir application and let mix take care of compiling and linking everything properly?

+1. That being said, universal k-12 schooling works because it is publicly run. A subsidized private sector model has a lot of bad incentives and issues to work out. As an example, I've sent my kids to a private school for the past five years, and last year our state introduced a voucher program to help subsidize private education. The school responded by raising the prices by almost the amount of the voucher, just for the age groups that it covered.

See also: US healthcare.

I have it too and feel the same way.

If I could easily just use it as a text terminal -- with emacs and ssh support -- I would use it every day. But when I looked up the hacking side of things it looked like a lot of work and kind of sketchy.

Just wanted to say your service is fantastic. We are using it extensively from our Elixir application so it's cool to see that make your list of languages -- is there an official integration coming out?

When I was most recently at Google (2021-ish) my team owned a bunch of SQL Pipelines that had fairly effective SQL tests. Not my favorite thing to work on, but it was a productive way to transform data. There are lots of open source versions of the same idea, but I have yet to see them accompanied with ergonomic testing. Any recommendations or pointers to open source SQL testing frameworks?

Oban is so good! My startup has an extensive graph of background jobs all managed by Oban, and it's just rock solid, simple to use and gets out of the way.

I'm in the same boat. I just cannot get them to work; they work sometimes on some browsers, but a solid majority of the time I click on "use passkey" I get a generic error message and end up going back and using the password flow.

I haven't invested more time in this because if it's so unusable for me as an engineer, it's a non-starter for the general public.

I've built document parsing pipelines for a few clients recently, and yeah this approach yields way superior results using what's currently available. Which is completely absurd, but here we are.

You can buy a RTX 6000 Pro Blackwell for $8000-ish which has 96GB VRAM and is much faster than the Apple integrated GPU.

Thank you so much for giving it a spin and reporting those issues :)

Also if you're interested, here is what your dashboard looks like (the onboarding flow is sadly the buggiest part of the app right now): https://imgur.com/a/eysQzjT

So the short answer is that this space is so early still that there’s plenty of space for more competition. But currently both of those market leaders are targeted at enterprise and have little to no penetration of the SMB/self-serve market. Ellm has a self serve onboarding flow, a $20/month monitoring only tier, and a $100/mo self-serve optimization tier — none of which require scheduling a call with a sales team.

I launched a product in this space to beta customers just last week -- https://ellm.co -- and the response has been way more positive than I could have hoped for. Every SMB owner I talk to is thinking about this and looking for ways to be ahead of the curve on it even though the number of commercial AI search queries is still dwarfed by Google. It feels like a race, and we are figuring out the rules as we run it.

FYI if anyone is looking for a production ready job runner in Elixir, I strongly recommend Oban. They have a paid pro version but the open source version is still absolutely fantastic.

Will payments to OpenRouter for hosted llama calls qualify?

Apologies since I have not taken the time to learn rust yet, but I've written a lot of modern C++. Is the ownership model kind of like std::unique_ptr and std::move, and `Rc<RefCell<Node>>` the same idea as `std::shared_ptr`? But less idiomatic? Or do I have the wrong idea?

This has not been my experience at all -- even as recently as last year it was

Recruiter: "Here's your offer for $X"

You: "Well I'm also negotiating an offer with Z for $X + 200k

Recruiter: OK well how about $X + 200k and a 50k signing bonus?

Multiple offers are key -- seriously, timing that properly is the biggest thing you can do to optimize your job search. Even if you can't give exact numbers, the mere presence of a competing offer will make a company offer you top of band.

I'm in loops at multiple FAANGs right now and it's not even that bad. Feels similar to when I did this last time? But I do get the impression that they're all just throwing out any resumes that don't already have a FAANG on them. If you're already in the club then it's ok, but I think it's harder than ever to join the club so to speak.

That's not true in the AI/ML space. But for everything else I would agree.

Apache Beam in Python does this, with code like

    counts = (
        lines
        | 'Split' >> (
            beam.FlatMap(
                lambda x: re.findall(r'[A-Za-z\']+', x)).with_output_types(str))
        | 'PairWithOne' >> beam.Map(lambda x: (x, 1))
        | 'GroupAndSum' >> beam.CombinePerKey(sum))

I'm not sure how I feel about it, other than the fact that I'd 100x rather write Beam pipelines in basically any other language. But that's about more than syntax.

Very different use case -- ZIPs/ZCTAs have some semblance of population normalization

Inbox is the one that still stings for me. I'm paying some silly amount for Superhuman which is almost there, but.. nothing is as good as Inbox was.

Or if your mail client, spam filter or anything else tries to prefetch the link...

This looks great; I've been working on something similar but I'll drop it since this is much further along. Do you have a list of issues/feature requests you'd like contributions on?