- cross-posted to:
- technology@beehaw.org
- cross-posted to:
- technology@beehaw.org
cross-posted from: https://lemm.ee/post/53487257
Realtime closed captions.
Arguable… it’s OKish at best, definitely nowhere near as good as professional… then IMHO it’s like spotting a spelling mistake in an official document, you instantly look for MORE mistakes then it become distracting. There is something powerful about trust that once it’s broken, it’s hard to get back. Once a spelling or here transcription mistake happens, then we brace for more (rationally so) and it becomes a very taxing endeavor.
So… sure STT progressed quite a bit but it’s STILL not good enough in a lot of cases.
Case in point, IMHO when there is a choice, most people (everybody?) would rather have human made captions than AI ones.
So for example if I want to add closed captions to my Twitch streams there is no way that I go pay a person to transcribe/translate the shit I’m saying for 3 hours everyday. Maybe this is a weird example but it fits my needs.
Not a weird example. I have my self hosted video server (PeerTube) and I tinkered with transcription thanks to whisper.cpp locally. It “works” in the sense that most of it is acceptable. It still does mistake though. I provide all my content, including hosting, at my costs and to anyone in the World for free.
So… I definitely see the value. I’m only saying that it has downsides and quality-wise relative to professional, it’s still bad.
Yeah, the song I 2 I, from a goofy movie will take AI a while to get right.
Absolutely nothing SAY IT AGAIN NOW
Pretty good for spam and scam.
Who’s Al?
What is AI good for - Images that are suitable for screen backgrounds, Creative text work - I had chatgpt work with me on a procedure to un-mothball a starship, it suggested details I did not think of. I had chatgpt also read and suggest re-wording and correction on text for documentation. I fed it a chapter at a time and proofread the output carefully to mitigate hallucination.
Here is the text above after a chatgpt re-write:
AI excels at creating stunning screen backgrounds and tackling creative text work. For example, I worked with ChatGPT to craft a procedure for un-mothballing a starship, and it offered valuable details I hadn’t considered. It’s also a fantastic tool for refining documentation—I had it review and reword chapters, carefully proofreading to ensure accuracy. The results were efficient, insightful, and elevated the final product.
reads like every corporate admemo. why sacrifice your voice? i preferred the original.
However, in the art arena AI is frustrating when you need a very specific image and not just a ‘a tree a dog a bird and a farmhouse’ AI art is getting better with text in the image but still makes some mistakes there. AI does poorly when you try to use it as a search engine : “What movie had a ball shaped robot” might bring back any answer. Warning though - AI is getting better at hands and spurious limbs.