TellowKrinkle

51 Followers
31 Following
132 Posts
@lina @Ember @disorderlyf Ahh. A lot of people were using ChatGPT as a translator because you can actually give it context, which can help in many situations. But it also just... does weird things, probably due to not being tuned for translation.

@lina @Ember @disorderlyf My big issue with LLM translation is that the increase in how natural it sounds is much higher than the increase in actual translation ability. Which means that the new translations seem much better while only actually being a little better, and people trusting Google Translate too much was already a big issue. While LLMs' accuracy has gotten better, I'm pretty sure the perceived accuracy / actual accuracy ratio has gotten worse, which isn't great.

On the hallucination side, DeepL is famous for turning single words into entire dictionary definitions, and both it and Google Translate can translate things entirely wrong or just remove pieces of sentences when they get complicated, but ChatGPT is the first I've observed to read into the source text and make up its own fanfic-like additional sentences to embellish the original text. And while that might be closer to the options that a human translator has when translating, I don't really trust ChatGPT to know when it should and shouldn't be doing that. So yes, while Google Translate did hallucinate, I also think LLMs' hallucination abilities are enhanced compared to Google Translate, and not in the "less hallucinations" direction.

RE: https://oldbytes.space/@gloriouscow/116224004520766154

There's a larger issue here here, and that is that it's trendy in certain spaces to be extreme and opinionated about your beliefs, and angry at anyone who doesn't share them. I see this a lot on Bluesky and Mastodon.

The problem is, this is a slippery slope towards ending up in a tiny bubble and losing many of your friends. And that doesn't lead to happiness or to good mental health. Not for you, and not for the people around you.

The two biggest topics I see this with lately is AI and trans discourse. The simple fact is, morality isn't absolute. Words don't have absolute meanings. Tools aren't absolutely evil or absolutely moral.

It's okay to be sad at the state of the world. I'm sad too! And it's okay to be angry at problem people (think, the billionaire class). But when you direct that anger at your peers, just because they don't share the exact moral compass you have, you're just hurting them and hurting yourself.

It's impossible to live in a world where your social circle is fully aligned with you on beliefs and morals. It just isn't. It's okay to be disappointed. But if you start cutting people off for it, you aren't making anything better.

(cont'd)

there’s many things which piss me off about modern web design practices but the worst one would be interfaces which show you a list or table of some small item and force you to look at this shit with pagination of like 50 items at a time

and then you try clicking previous and next to actually find whatever it is that you’re looking for and pray to all gods that this thing isn’t too dumb to remember where it was and is going to just return some random items from the middle

motherfuckers will pour megabytes worth of javascript down your pipe, but sending more than a few kilobytes worth of actual payload is too much bandwidth for them

anyway this rant brought to you by my attempt of looking at a PR review on github and being stuck trying to figure out how to load more than 40 comments at once

you know what is good at handling large amounts of text at a time? computers. you should try it sometime. it’s kinda amazing how much text you can fit in a few hundred megs of RAM.

To exist online in 10 years, you're going to have to give a random company a face scan and photo ID and they're going to give that and everything you post to the government to spy on you for being queer, left-wing, "a criminal," or whatever. None of your data or communications will be private. The government and corporations will go through all of it.

If you don't want that to happen, you should raise a stink about "age verification" right now because that's what they're actually building.

We already have a good, entirely client side method of verifying whether people are adults. It's called "parental controls".

Add an API for websites to query it. Maybe add some content warning meta tags to html headers for the browser to automatically block pages if they shouldn't be allowed. This should not need server side verification.

Please stop asking people to upload their IDs to the internet.

@dougall Good to know, I didn't get an M1 until they had been out for a while so I guess I missed all the fun. And yes, I've filed with feedback assistant.
Has anyone else had trouble connecting M4 MacBooks to monitors?
My M4 Pro frequently fails to connect to an Asus PA279CV over USB-C (power and devices on its USB hub connect but the laptop acts like no monitor has been connected).
Weirdly, connecting or disconnecting a power source to/from any of the other ports on the laptop (USB-C or MagSafe) while the monitor is connected fixes the issue.
Reproduced with a coworker's M5 as well.
@whitequark oh yeah if you do happen to want to buy the LN or manga, the English publisher (J-Novel Club) sells DRM-free epubs on their website which I love. If only all the other publishers did this too...
@whitequark if you're using fish shell, it can include some short explanatory text next to the options in the autocomplete UI. It's very nice when available (but not super common).