self

the word promulgating is creaking under the strain of this sentence

ah yep, this is for Spanish! I’ll give memrise and that languagejones video a look

I think what excited me about Readlang is that I could use real text (which gives my ADHD brain something other than language learning to be excited for) and it’s like a semi-automated version of the process I’ve seen Spanish speakers use to learn English with a translation dictionary and handwritten flash cards. I might just have to try a variety of apps (for as much as these free trials will let me — a lot of these companies are terrified I’ll learn something of value without paying them a fuckload of money) until I find one that tickles the same part of my brain

can anyone recommend a language learning app or system that isn’t dependent on LLM garbage? after bouncing off of Duolingo I almost landed on Readlang, but:

• all of its features seem to be LLM-dependent
• because of that, its word and phrase explanations have a bit of oddity to them that I feel will get worse when I get past the beginner material
• it’s a lot slower than it should be because it’s calling into ChatGPT for everything
• even though this is supposedly their strength, I’ve had really bad instances where a GPT-based translation app translates Spanish (which should be fairly easy) into absolutely nonsensical English, and I’m kind of terrified I’ll make a fool of myself learning Spanish from a system where that’s a statistically likely probability
• maybe I don’t want to pay some asshole to not write me some study materials????
• plagiarism and the rainforests
• it feels like Readlang really doesn’t need an LLM or a $6/month subscription that’ll almost certainly go up? like, it’s essentially an e-reader with a manual translation feature (that could be just a Spanish word/phrase dictionary) that also generates flash cards whenever you activate the translation. is there really not an e-reader or browser plugin that just does this shit without LLMs?

with that rant out of the way, I’m open to suggestions that aren’t Duolingo’s model or another round of passing grades and zero vocabulary retention at the community college

this might run afoul of some of our scattered price gouging regulations (see some instances of sellers getting slapped during the toilet paper shortage nonsense), but interestingly this was announced after the Supreme Court kneecapped our ability to enforce any regulations at all

it’s almost like the corpos took a monstrous lesson from covid and were waiting for the right combination of deniable technology and probability of the success of the Supreme Court’s judiciary coup to announce something like this

• don’t use any of this stupid garbage
• if you’re forced to deploy this stupid garbage, treat RAG like a poorly-secured search engine index (which it pretty much is) or privacy-hostile API and don’t feed anything sensitive or valuable into it
• document the fuck out of your objections because this stupid garbage is easy to get wrong and might fabricate liability-inducing answers in spite of your best efforts
• push back hard on making any of this stupid garbage public-facing, but remember that your VPN really shouldn’t be the only thing saving you from a data breach

no matter how you hold it, you’re holding it wrong:

"It's kind of funny in a way - if you have a bot that's useful, then it's vulnerable. If it's not vulnerable, it's not useful," Bargury said.

it’s excellent if you can get it from a street taqueria!

we need to bring back machinima

I call dibs on being represented by the left pong paddle, lovingly recorded from the analog output of an Atari Pong TTL standup (or FPGA equivalent)

Why Choose AI Pastor?

usually because I’m tired of carne asada

wow, remember when a bunch of random posters came to that security thread to try and gaslight us into thinking the very similar attack described in @dgerard@awful.systems’s blog post wasn’t a security vulnerability? and now it’s a Black Hat talk, aka “you fucked up and now the world knows about it”

"It's kind of funny in a way - if you have a bot that's useful, then it's vulnerable. If it's not vulnerable, it's not useful," Bargury said.

holy fuck that’s damning. LLMs are so worthless on their own that they can’t do anything unless you’ve got everything hooked up to RAG, which is just a wide-open API with access to all your data.

I like the idea of small communities, but a major issue (possibly the biggest issue) as demonstrated by many mastodon servers over the years is longevity. What happens when your admin gets bored/burns out/dies/goes fash/is replaced with an asshole/is unable or unwilling to moderate effectively?

this is something I’ve been thinking on quite a lot myself — how do we (being a small web service without effectively unlimited VC money to burn on cloud credits or an entrenched corporate infrastructure) have continuity in case anything happens? and as an established community, that continuity has to encompass our infrastructure, our data, and the understanding and expectations that make moderation work.

• for infrastructure, we’re somewhat ok — our deployment code is open, and there’s just enough docs that a replacement admin can spin up an identical cluster with a bit of work
• data’s a lot harder. I’d love to regularly publish a dump of our database with the sensitive details redacted to as many places as is practical (there’s a bunch of archive sites for this), but that would open us to a number of garden-variety and lemmy-specific attacks (and I won’t be describing those in public for obvious reasons, but established posters can inquire in DMs). most likely in the short term this’ll involve rsyncing full database and image storage dumps to trusted parties on a regular basis, though I’m open to any better ideas.
• the problem of guaranteeing continuity of moderation is unsolved. the only idea I have in this direction is effectively a guild or co-op model that’d exist to teach and certify moderators and admins how to maintain communities like ours. I haven’t taken any steps in this direction, and there’s a lot to the idea that’s still effectively magic (how should certification work? what systems should be in place in case of bad actors? should this thing itself be a mostly technical solution or a mostly social one?), but it could potentially guarantee moderator continuity for federated systems other than ours too.

my thoughts on paully’s output are along the same lines — it’s really a shame that he’s the guy a lot of people learned Lisp from, because it’s very clear he’s just a rich white guy bullshitting on that and every other topic he’s famous for

also, this stood out from a quote in your first linked post:

Server-based deployment of software was a central theme in Graham’s essays, and his continuation-based web framework was an interesting and fairly novel way to create continuity across multiple requests in a single session.

Seaside did this first, though it never got a ton of traction. if naive continuations are how the orange site’s doing sessions and state tracking, that goes a long way towards explaining why it’s so incredibly bad at scaling and has such weird performance characteristics. there’s ways to make continuations more performant in this role, but it takes a degree of low-level understanding paully’s never demonstrated, since his languages have always been built on top of Racket (which is a fine language for making languages! it’s fun as hell! but one day you will run out of runtime to repurpose)

(I should see if Racket finally has good LSP support for #lang languages. I’d seriously use it so much more if I could bring my own editor)

also, holy shit those responses to your second linked post really haven’t aged well at all, have they?