IAAP :
Transcript
I think when I think of some other examples of, that sort of automation with and without
human in the loop. I think one of the biggest examples is maybe even closed captions,
right? You can go to really any YouTube video or even Zoom or Google Meet and hit the
closed caption button. And it's some percent accurate, right? I think for every 1% of packet
loss, you know, you lose maybe five or 10% of understandability is the numbers that I've
heard. Packet loss being, you know, 100% accurate translation versus 99% translations
and 1% packet loss. So you would assume from going to 100 to 99, you would only lose 1%
of its understandability, but it's actually much more severe. So even if a tool is 95%
accurate at translation, you might be losing 40% of the actual understandability of it, right?
So that's a really interesting nuance, but I think the same can be applied to alt text, right?
Your alt text might be 99% perfect, but that extra 1% is gonna come through that human in
the loop who's gonna approve a change or something. And so we always think how do you
reduce packet loss in addition to things like how do you reduce false positives
2026-07-01 13:50:43