Dan Conway (@magisterconway.bsky.social)

So, um... this is bad. Really bad. I looked at the letters that were translated by the AI, and the very first one I found was almost entirely hallucination. Thread: [contains quote post or other embedded content]

Blacksky

"AI" users are like, "I know this is imprecise but as a convenience these transcriptions are better than nothing"

then 70 years from now we'll still be struggling to debunk these entirely hallucinated transcriptions of thousands of manuscripts that were pissed into the pool of human knowledge.

some things are worse than nothing. "signal-shaped noise" is worse than nothing.

@elilla I mean there is a reason regulations around captions exist. In the US they’re usually required to be accurate. It used to be you could see real-time captions being backspaced on TV sometimes.

But nobody has enforced it for years… AI or not, modern captions are *awful* because they’ve outsourced their production, often to underpaid second-language speakers, or the outsourced contractors often just use AI too.

In the US, it used to be that NCI and VITAC did them, and were extremely good at it.

@elilla tangential:

I’m not shitting on second-language speakers here but captioned besides being human, should be competent at a near-native level in the language of the source material and the captions to do captioning.

Very often they just aren’t now because… they wanna save a few thousand bucks on something that costs millions to produce to begin with. 🫩

Speaking from my native language, English is phonologically tricky and it’s easy to make a malapropism like “kill two birds and one stone” or “for all intensive purposes.” Native speakers even fuck it up sometimes.