Mastodawn

Robey ☠️

@robey@messydesk.social

490 Followers

161 Following

6.3K Posts

A wizard appears. He looks grumpy.

I was the guy behind eggdrop, paramiko, the sidekick ssh client, kestrel, and many smaller projects. I've been in 3 unsuccessful bands and I'm one of the 23 people who finished reading "Gravity's Rainbow". But I'm primarily known for making a dumb Fibonacci joke online in 2011.

$argon2id$v=19$m=512,t=256,p=1$uksdy+s78ZlpzoaxkGyciA$kIt70t1MNJ5auxeTAohzAMKBec73DvmFYct+o8MrQy4

#programming #retrocomputing #firmware #leftist #content

Location	San José, California
Disposition	gay (they/he), pale, quiet, antifa
Nerd topics	distsys, firmware, retro computing
Blog	https://robey.lag.net/

Robey ☠️22h ago

@aphyr i think an artist could do a lot with dapper eugene v debs

Robey ☠️22h ago

Kristina Gehrmann 1d ago

Forest Patrol, or: Frogbert's first hunting expedition. 😆

(Personal work)

Robey ☠️22h ago

decryption 1d ago

There's a guy on TinkerDifferent that got their hands on an early prototype of the Finder from 1982 and some other early demo apps - including the Bouncing Pepsi Caps (made to impress mike scully) app.

MacPaint 0.1, the IconEditor and a few others. This is all stuff the Mac team was using to build the Mac. Pretty cool! I don't think any of these apps were outside Apple until now?

https://tinkerdifferent.com/threads/early-demos-finder-soundlab-sbardemo-windowmanager-iconeditor.4498/

Robey ☠️22h ago

DJ Sundog from the *new* toot-lab 1d ago

when I was a kid, the only job I wanted to end up with was to be a professional musician.

when I grew up and decided that I wasn’t good enough for that to happen (editor’s note: I was totally good enough for that to happen) I landed on my computer experience and turned that into employment.

it went well enough that I kept going down that path and ended up working in the very companies that were making computers and networks happen.

now, having left that industry a couple years ago, I still do computer stuff. network stuff. probably always will.

but I don’t see any path that would ever lead me to willingly go back into the tech industry proper. not in any semblance of the form it now takes, at least. maybe if the whole current crop of computer corps foundered en masse and the vacuum were filled by a bunch of small teams building delightful tools for their customers I’d think about it. but honestly, I’d always be suspicious, I think.

so many folk with similar stories these days. the number of unrecognized self owns the titans of tech are committing are just astounding.

Robey ☠️22h ago

Speak for yourself, meme

Robey ☠️22h ago

@webuiltthiscity dare they challenge our poop serpent?!

Robey ☠️1d ago

tried the new thunderbird and find the ux very frustrating. that's too bad since i think i read a blog a while ago where they did a big redesign.

when setting up a basic imap account, every field i touched would modify a different field that i had to go back and fix. like an evil puzzle game.

also every received email gets an "archive" button which would be useful if it did anything, but it seems like they forgot to hook it up.

Robey ☠️1d ago

well, well, well... if it isn't my old friend systemd/journald...

Robey ☠️1d ago

the moon keeps going through its phases... the rains come ever winter... and once a month mozilla reminds us they know absolutely nothing about their own user base

you can set your watch to it

Robey ☠️1d ago

×

@brooke I kinda like the on-device Translations feature 👉👈

(but this one is stupid)

@brooke although it feels notable that the models for Translations are not *exclusively* the work of Moz, so maybe the outside contributions help get it over the bar of "some c-suite dipshit's idea" and into "actually useful" territory

Stephen K 🔑☢️⚛️🇺🇦

@SnoopJ @brooke Yes but on device translation can be done without AI, has been done before via OCR and simple libraries.

@CommieGIR I am not aware of any "simple libraries" that achieve machine translations of the same quality as the recently launched EN<->JA language pair. If you are, I am interested to hear specifics.

And OCR has nothing to do with it: the text is right there in the page already.

Stephen K 🔑☢️⚛️🇺🇦

@SnoopJ Having had to translate a lot of Manga - There is absolutely tools for doing the translation, JA to EN is just just a matter of availability that makes LLMs attractive

The reason LLMs are popular for this is they are chatbots, so they just fit.

But it was being done long before LLMs were a thing.

Most of them are just scraping the text and running it through a non-LLM translator service, the LLM just handles some of the nitty gritty work

@CommieGIR as I said, if you have specific recommendations about such a tool that I can use in my browser, I am open to hearing about them.

I am aware that LLMs are not the first tools to do this, and nowhere did I state otherwise. I would describe NONE of the preceding tools I am aware of as "simple".

Stephen K 🔑☢️⚛️🇺🇦

@SnoopJ https://github.com/LibreTranslate/LibreTranslate

Was one I used for a lot of early Manga transcription, but used an OCR tool to scrap the text off images.

Its based off OpenNMT: https://opennmt.net/

Kyle Lacy 18h ago

@CommieGIR @SnoopJ i don’t want to be pedantic but transformer-based LLMs were literally built for machine translation originally (by researchers at Google). I think OpenAI were the ones to take that architecture and used it to make a generative model (“transformer” is the T in GPT)

OpenNMT itself is also an LLM, I’d argue. it is a language model, and the pretrained models are up to 3B parameters (similar size to self-hostable LLMs like Llama). only difference from LLMs in the modern era is it’s based on LSTM rather than the transformer architecture

I don’t think there’s a line in the sand where you can say “OpenNMT is machine translation (good) and Firefox’s translation/summarization model is an LLM (AI) (evil)”. it’s the same idea but using a newer architecture with (presumably) better results

edit: forgot the original post was about summarization not translation. IMO basically the same stuff applies (OpenNMT can do both too)

@kyle @CommieGIR other machine translation methods did exist beforehand, but many of them had notable flaws. Particularly long-distance correlations.

It's a novel approach but certainly not the only one. I do think the per-pair models are quite a good approach inasmuch as the problem domain is limited and the architecture can shine. Same dataset woes but those are table stakes anyway

@brooke ???
I turned it off as soon as I discovered its existence, so I hadn't actually seen it in action, but… WTF?
I thought it would just pop up the normal "summary card" that we're familiar with (containing what the site author explicitly decided to show), not a stochastic slop!

@dakkar @brooke Yeah the browser market is essentially lesser evil choice now. Sure you can go for some super niche hyprland optimized browser that will burn your computer after every use… but from big three I still pick #firefox and yes #mozilla pushes weird stuff but at least there is a way to disable it.

@blami @brooke no argument there! I run Firefox on all my machines

@brooke AI generated?

@Lioh some people know how to use image editors still :)

@brooke All in all, their decissions pointed me in direction of the Zen browser which took me a couple of days to get used to but now I love it. And yes, is based on Firefox but no, has none of the AI things I don't want it to have.

Aaaaand rescued Evolution as a mail client which nowadays is pretty good and makes not miss Thunderbird.

So... good job, Mozilla.