Mastodawn

[object Object]May 27

I’m glad people are finally noticing LLM translators will just make plausible sentences the fuck up when they’re fed anything but a perfect source to translate, which makes them exhausting and damaging for language learning and a variety of other situations where you’re expected to actually, you know, use the fucking language for anything but a highly inaccurate skim

thank fuck all the language learning companies didn’t jump onto the LLM train, right?

…right?

[object Object]

this post brought to you by the language learning app I’m using semi-consistently telling me that names, formatting artifacts, and words in languages other than the one being learned, all have fabricated definitions related to the sentence they appear next to

this is a very hard error to catch if you aren’t paying attention and it’s actively slowing me down

[object Object]May 27

“LLMs might be horrible at almost everything but at least they’re good at translating and accessibility”

are they though? or are they fucking terrible and you didn’t notice because the output looked good and you didn’t check it?

@zzt LLMs: good at everything that you are not equipped to verify.

@ainmosni @zzt ai is great at things that are very cheap to validate, because it learns to get the person that gets 3¢ per validation to click the green button

@kabel42 @zzt sadly, while the validation is cheap, the generation isn't, but that's not their problem.

@ainmosni @zzt Is it? I can't validate a translation is good that fast, I cant even decide which squares are bicyles and not mopeds

@kabel42 @zzt I meant in cases where validation is cheap, like the insane tightloop in claude code that validates if claude generated valid json with an open jsonschema and then lets claude try again if it doesn't validate.

@kabel42 @ainmosni @zzt
> [ai] learns to get the person that gets 3¢ per validation to click the green button

in this setup, ai learns to produce plausible text rather than correct

Flesh & bone by the telephone May 27

@zzt LLMs are always really good at whatever type of output you don't know how to validate.

If you don't know how to code, it is great at code.

If you don't know a language, it is great at translating it.

If you don't know the law, it is a great lawyer.

Every fucking time... Yet somehow, some people aren't seeing the fucking pattern.

The Penguin of Evil May 27

@minego @zzt Exception - I know how to bullshit managers, it's great at bullshitting managers

Flesh & bone by the telephone May 27

@etchedpixels
@zzt It's great at bullshiting managers who don't understand the work their employees do.

My manager used to be a coder, and he gets it. He sees how bad it is, and listens. Sadly his manager doesn't.

@etchedpixels @minego @zzt

Ah, is that an exception, or just an extension of the principle?

LLMs are great at whatever output you don't know how to validate.

Managers and C-suites generally don't know anything about anything.

Therefore, they consider LLMs universally effective and highly desirable.

Lord Caramac the Clueless, KSC May 27

@TheEntity @etchedpixels @minego @zzt Why don't we just replace all the managers and executives with LLMs then?

The Penguin of Evil May 27

@LordCaramac @TheEntity @minego @zzt Certainly for the management consultants I'm not sure anyone would notice

F4GRX SÃ©bastien May 28

@etchedpixels @minego @zzt including tender responses.

@etchedpixels @minego @zzt Which isn't that hard, just phrase it in that weird corpobollocks pidgin language they speak and they'll be convinced.

The Penguin of Evil May 28

@nini @minego @zzt 'embracing the latest upward debt leveraged technofuckery'

Or you can use AI against itself as Kagi translate will happily let you set the other language to anything including linked in which is a hoot - or even perl

Roger BW 😷May 27

@minego @zzt I stopped reading The Ecnomist when I noticed a similar pattern.

Marjane Strapi 🍆 🥺May 27

@RogerBW @minego @zzt
rare to see gel-mann amnesia outside of journalism, but there we are

Claire McNab May 27

@minego @zzt The solution to that problem is already being implemented: fire the people who know how to validate.

Alison Wilder May 27

@minego @zzt hence the reason people whose only job is to tell other people to do things think it's good at everything

@minego @zzt llevo toda mi vida teniendo esa misma sensación con el periodismo. Todo parece verdad hasta que dominas el campo del que trata el artículo y entonces todo parece estar mal. Mal entendido, mal explicado y tendencioso. Así que no voy a notar tanta tanta diferencia. Seguiré no creyendo casi nada de lo que me digan y aplicando siete capas de escepticismo.

@minego @zzt I think it's called Gell-Mann Amnesia.

It's the same thing with the warehouse robot, by the way. If I worked the speed that robot (or his human worked), I'd be fired. If I worked in conditions a tenth as simple, uniform and reproducible, I'd buy myself a lottery ticket.

rob los ricos May 27

if you are not into hallucinogens, it is great at hallucinating.

@zzt I translated something to English yesterday, and the translator made up an “English” word

fuzzyfuzzyfungus May 27

@scm @zzt That's...technically...in line with how english-speaking humans handle english.

If we aren't under pressure to be 'generative' we often just yoink the other language's word and call it good.

I suspect that the bot is not driven by a commitment to idiomatic perfection, however.

@fuzzyfuzzyfungus @zzt it wasn’t a word in any language, as far as I know, I think it was a mangled English word

Jeff Grigg May 28

You resist its work to embiggen your vocabulary with a perfectly cromulent word!

😉

Matt Mascarenhas May 27

@zzt So true. Also related, I reckon, is how they groom these same people – those who haven't tried doing the actual task – into thinking the task to be easy / below them.

Like, people who don't bake or dance, watching Bake Off or Strictly, seeing people make mistakes and judging them super harshly for it.

Keeping the users from checking the output and trying to do the task themselves, is intrinsic to the con, imo.

Author Andrew Massie 🏴May 27

@zzt A week or two ago I had one of those funny coincidence moments where I saw a post like "It's sad that AI companies are doing this because machine translation is one of the only things LLMs are good for," followed about an hour later by "I had to retranslate this entire passage because the machine translation was nonsense."

It's almost like LLMs aren't good at anything

Tael 🔜 AC26 May 27

@awmwrites @zzt Come now, let's be reasonable. They're good at generating intelligible strings of text in many different languages. Otherwise, we wouldn't be in this situation.