@mcc the RoI of learning Chinese to read poetry (at which machine translation tools suck) is very different from RoI of learning Chinese to read another stuipd datasheet for something you pulled out of e-waste (at which machine translation tools are quite good or at least sufficient because it is so formulaic)

source: I have tried the machine translation tools and they were the shit

@whitequark @mcc I hate the cliche of "word calculator" but I think machine translation definitely fits that bill when you just want a sense of what the text says (rather than the ability to read it as a native)

I think part of the reason the field has survived the rise of slopware is because it existed long beforehand, so "a worse way to do it" wasn't as attractive of an option and specialist models (Bergamot) are still dominating that use-case.

I do think a lot about how one of the models popular with Enthusiasts once gave me a very detailed explanation of why 私 is a more polite "word" than わたし. Very accurate reproduction of a guy from the 2000s who knows absolutely fucking nothing about Japanese.

@SnoopJ @mcc yeah. if I want a translation that is actually good I will simply pay a human translator (something I've done before!) for a result incomparable with either machine translation or learning the language myself. but I would only do this for source material I like. I engage a lot with source material that is adversarial: where it only exists as an obstacle towards something I want (e.g. turning e-waste into not e-waste).

There is absolutely no way I will grace such material with hundreds of hours of my lifetime.

@whitequark @SnoopJ @mcc I agree that formally this is true but colloquially, “is this person spending effort” is much easier to evaluate than to be able to deeply interrogate the quality of the results of that effort, which is why the expression is legible to me.

this is one of the most frustrating aspects of LLMs; they are spam machines, that spit out not just bullshit but *superhumanly plausible* bullshit, which means that reasonable heuristics for evaluating output quality malfunction

@glyph @SnoopJ @mcc I think you are calling out something important and relevant here, which is that we live in a society that attempts to judge people's interiority:

• is this person actually ADHD, or are they just lying about it for the drugs?
• is this person actually a woman, or are they just lying about it for the access?
• is this person actually mentally ill, or are they just pretending to be for the attention?

I see...

• is this person actually putting in effort, or are they just lying about it for the social approval?

... as a natural addition to this corrosive series. I think being in an adversarial relationship with the interiority of strangers is one of the worst ideas we have normalized: even when it "works" it fucks up people for life. "I must be a fraud because I don't feel like I put in enough effort" is both a common and a deeply awful lesson to learn. I'm sure you've encountered people who learned it, so ask them, how is it going?

(I have heard that some other societies, like China, put more social emphasis on performance of the ritual than on performance of the belief; as in, you can believe whatever you want but you better act as-if you believe the things we reward. that has its own problems but I much prefer it to the above, with the caveat that I don't know how accurate of a representation this is.)

lastly, I don't think "is this person spending effort" is at all easy to evaluate. is the effort of a person who finds learning Mandarin easy intrinsically more valuable than the effort of a person who finds learning Mandarin hard? given the same result for both I'd say no, but since both pay for it with their lifetime, under the proposed evaluation rules the answer would be yes.

quiz: do you think I, as an open source developer, "put in the effort?" (this is a sincere question)