irskep

It's not really true to say "very little has been done." Thousands of person-hours have been invested into this problem! But the results have been mixed.

At least uv is nice! https://docs.astral.sh/uv/

> Any solution that's 80% accurate just doesn't work for most applications.

And yet people use LLMs, for which "80% accuracy" is still mostly an aspiration. :-)

I think it's reasonably likely most people companies end up using open source libraries, at least partly because it lets them avoid adding another GDPR sub-processor. Unstructured.io, one of your competitors, goes as far as having an AWS Marketplace setup so customers can use their own infrastruture but still pay them.

LLMs might get better at consuming badly-formatted data, so the data only needs to meet that minimum bar, vs the admittedly very nice output you showed.

The only reason I'm not immediately answering is because I need to check whether it's a trade secret. We do our own thing that I haven't seen anywhere else and works super well. Sorry for being mysterious, I'll try to get an OK to share.

Edit: yeah I can't talk about it, sorry

I worked on an in-house version of this feature for my employer (turning files into LLM-friendly text). After reading the source code, I can say this is a pretty reasonable implementation of this type of thing. But I would avoid using it for images, since the LLM providers let you just pass images directly, and I would also avoid using it for spreadsheets, since LLMs are very bad at interpreting Markdown tables.

There are a lot of random startups and open source projects who try to make this space sound fancy, but I really hope the end state is a simple project like this, easy to understand and easy to deploy.

I do wish it had a knob to turn for "how much processing do you want me to do." For PDF specifically, you either have to get a crappy version of the plain text using heuristics in a way that is very sensitive to how the PDF is exported, or you have to go full OCR, and it's annoying when a project locks you into one or the other. I'm also not sure I'd want to use the speech-to-text features here since they might have very different performance characteristics than the text-to-text stuff.

I'd love to see this hooked up to Navidrome, Jellyfin, or one of the other home server music solutions. Such a vibe.

This looks like a list of cross-platform command line tools. What about it is Mac-specific?

I did say "at risk of" - the front page of HN is a powerful place. :-) Looks like my fears did not come to pass.

The title is sensationalized. I read the whole thread. Gitea doesn't seem to be "blocking" the PR. They aren't preventing it from going in a release. However, they did pull the in-progress PR into their private repo early, presumably in order to make more money. Not great, but not bad faith.

"Blocking" would suggest to me that the PR is rejected but still ends up in the closed-source codebase, which isn't the case at all.

@dang I flagged this because we're at risk of brigading a community over a nothingburger.

Taking a shot at a gentler version of the sibling comment.

If someone says "I made a thing," and a reply says "that's cool but I would rather have something slightly different that matches my technology of choice," I can't see any positive outcomes for anybody. For the submitter or author, the message is "what you made isn't good enough." For a reader without context, it purely muddies the waters because it adds noise without teaching them anything. For a reader with context, it's not new. I don't understand what the value could be for anyone.

Imagine someone submits an article about a MySQL ORM for Go, and the top comment is "now one in Ruby, please." How does that help anyone? It only diminishes the efforts of the submitter and ignores the complexity of writing an ORM in the first place.

If you want to start an effort to do something similar, then talk about doing that. If you want to tell people about another cool technology, do that. If you think the project under discussion could be extended to support something you want, pitch the value. If you want to make a feature request, you can definitely do that with more kindness for the creator.

I know from experience how annoying it is when random people come at you with tone critiques to your hastily written HN comments, so please take this as intended, just an attempt at building empathy.

I've used both extensively. (Material for) MkDocs is easy to set up and gives very good UX for users. Sphinx is hard to set up and provides better UX in some ways (internal links are on another level entirely), and worse UX in others (the themes are all worse than Material for MkDocs, every single one).

I really appreciate that in their big list of problems they've seen, they also discuss the solutions in detail instead of just dunking on these places.

Is it also supported in the Google Drive export API? It's not in the docs yet at least. https://developers.google.com/drive/api/guides/ref-export-fo...

I wrote a web site builder that I only use to host my indie bands' web pages. It's got a very silly backend vs the potential target market (musicians) in that it makes you write Markdown, so I don't have delusions of grandeur, but it works well for my simple hosting needs and lets my bandmates make changes without going to git.

I'm predisposed to agree with the diagnosis that incompetence is ruining a lot of things, but the article boils down to "diversity hiring is destroying society" and seems to attribute a lot of the decline to the Civil Rights Act of 1964. Just in case anybody's wondering what they would get from this article.

> By the 1960s, the systematic selection for competence came into direct conflict with the political imperatives of the civil rights movement. During the period from 1961 to 1972, a series of Supreme Court rulings, executive orders, and laws—most critically, the Civil Rights Act of 1964—put meritocracy and the new political imperative of protected-group diversity on a collision course. Administrative law judges have accepted statistically observable disparities in outcomes between groups as prima facie evidence of illegal discrimination. The result has been clear: any time meritocracy and diversity come into direct conflict, diversity must take priority.

TL;DR "the California PG&E wildfires and today's JavaScript vulnerability are all the fault of Woke Politics." Saved you a click.

I wonder if a lot of the heated discussion is purely an emotional reaction to be hit with "AI feature onboarding" immediately after a normally-drama-free terminal emulator software update. If they had gone with a much more low-key way of notifying users of the feature, maybe folks would be more reasonable.

Not that I blame the maintainers for using the approach they did, though! Usually you do want to be super obvious about new things.

Ehhh...I can see how people interpreted it this way based on the way the onboarding screens looked. It was very jarring to open up my terminal one day and be greeted by yEt AnOtHeR gReAt Ai FeAtUrE. I imagine that emotional reaction is fueling the discourse quite a bit.

This is an unfortunate failure of community engagement. Apparently the feature was in beta for months, but given that most people don't use the beta or follow the development of their terminal emulator, it shouldn't be a surprise that there was pushback after release. Usually when introducing a controversial change, you want to do additional community outreach.

It's also a minor UX failure. The onboarding screens weren't loud enough about the need to provide an API key.

I really feel for the maintainers here. They added a cool thing to their popular-for-a-reason terminal emulator, and (from their perspective) people lost their minds over a misunderstanding or (again, from their perspective) unreasonable expectations.

Those words were definitely written through gritted teeth, but it seems to me he's being very literal. The second point in particular is exactly what the anti-AI-feature people seem to be asking for (and they are reasonable for asking for it).

It's literally a Tinder-style swipe interface for musician matching. I recommend taking a realistic look at your competition.

How does this differ from Vampr and Bandmix, which claim to be the same thing?

I can't say it's "any level of scale," but I used this for https://diyband.net and it was very straightforward for the needs of a hobby-scale web site.

The pitfalls are:

1. It's SQLite, with everything that entails. The schema migration story is "SSH into your production machine and run your migration scripts."

2. Every deploy to a new machine involves downloading your database from S3. Zero-downtime deploys aren't possible because you need to pause writes, sync, and then download the DB to the new container or whatever. If your database is gigabytes (mine is kilobytes), this could be a deal breaker.

Honestly, if managed databases were as cheap as SQLite hosting, I'd just use a managed database, because the SQLite ops workflows are too weird. But I'm confident in the reliability and performance of the site, and it is genuinely cheaper to not run a managed database, which is important for hobby projects.

(Standard internet comment disclaimer that I am a fallible human who might have missed something or am somehow doing it wrong...)

It actually wasn't WWDC, it was an interview some execs did with John Gruber, which unfortunately was 2 hours long. Segment on visionOS starts here: https://youtu.be/DgLrBSQ6x7E?t=2323

I don't have time to dig in and grab the exact spot, but I'm pretty sure they didn't go into detail much beyond what I said. My guess is text rendering is as good as it can be with state of the art displays, from a software stack standpoint.

Edit: I would have loved to add to the conversation, but people do love their pitchforks and I should be working instead of commenting on HN. :-)

Admitting you worked for Yelp on HN: not even once

In one of the WWDC videos, they mention they reimplemented their glyph renderer to be vector-based, so text should at least not depend on first rendering to laptop-screen pixels and then to face-screen pixels. Seems like it should be state of the art at launch.

I love that everybody commenting here here emphatically insists that you must be doing it wrong, but real-world solutions are nowhere to be found. I admire your commitment to keeping a positive tone, but you'd be justified in feeling some frustration.

I do hope a PR shows up in the next few days so the internet can stop having this debate and move on with a good library.

You're right, the code is not performant at all. At the time I was hacking around and just happy to get something working. I think the code is a tempting distraction, but I was trying to say that browser API performance is not always what you would expect and sometimes you need to get creative in order to get the results you want.

I think people in this thread are being much too optimistic about browser performance. Writing a loop isn't the issue; memory access patterns are. Canvas implementations are just not well suited for the flood fill problem space. (I spent some time with flood fills + canvas ten years ago when I was working on the now-dead library Literally Canvas, but I assume that experience is pretty far out of date now.)

Precomputing fill areas is a really good idea for the situation you're in!

Edit: actually it was 7 years https://github.com/irskep/literallycanvas-pro-tools/blob/mas...

My worry with this kind of thing is that some random dependency will break and there will be no easy solution. Compiling Python to JS seems like a very fragile process. And if people run into bugs, they will need to debug a very specialized and unusual toolchain.

If you're going to go all-in on one language to build an entire SaaS product, JS seems like a much safer choice. There's no Django-for-JS yet, as far as I know, but when there is, it'll be a no-brainer.

Looks like it was just a different link that somehow beat the other one to the front page for a while. https://news.ycombinator.com/item?id=35150345

@dang, it looks like all comments from an hour+ ago on this article have been deleted. They were OK comments, is there a bug?