I understand why people are using AI to help create. Especially if you produce content instead of art. Production for work (content) takes forever and is a joyless process, whereas production for art is usually fun and healing. Production for work is supposed to lead to money (always).
I very much prefer human-written, human-filmed, human-recorded, and human-edited content. It’s still so much better than AI created content. As soon as I sense the AI in a video I hit skip/next.
But what is absolutely infuriating and makes me wonder what’s really going on is when the AI voiceover mispronounces a name or word. HOW? WHY? Often after correctly pronouncing the same word. Usually it’s not an obscure or foreign phrase mentioned once, but the main subject of the video!
I’ve seen videos where the main subject of the video is mispronounced several times in the same video. As a developer and a producer, how does this happen?
My developer brain —-
How could the AI decide to pronounce the same word differently each time? Does it have any idea what it’s outputting at all? There’s nothing about computer programming logic that results in pronouncing the same word differently if the output is set for a single language.
Why would the AI even know incorrect pronunciations at all? A few words do have an alternate pronunciation (tomato), and there are even some regional dialects built in, but that’s different than completely incorrect pronunciations. If there are options, wouldn’t the AI either pick one and use it every time, or even explain the differences between them?
My producer brain —-
Hey I’ve been working with this AI to help make videos about Topic A. I’m going to have it make some content about Topic A with great images of Topic A and 10 minutes of info on the history and current market status of Topic A. I’ve done the research and written the prompt on Topic A, and the AI just spit out a test video on Topic A. NICE.
Do you think I have 10 minutes to watch the output? Do I even notice or care that Topic A was referred to as Toe Picka twice, Topic Ah once, Too pick-uh once, and even Top I.C.A. once?
Any human anywhere would catch these errors. But AI does not. Even playing the audio would catch these errors.
What going on here?
It’s just AI slop that no one edits, no one reviews, and no one cares about. Who knows if a human even prompted it. If they can upload that video and make more than a few pennies from Youtube it’s worth it. There’s probably not a person watching any of it, or even uploading it. It’s all automated. SLOP.
It’s called “content farming” and it’s definitely not new, but AI has made it more popular than ever. It’s now crowding out human-created content on many platforms.
But if you want some conspiracy theories I have a few. Just for fun, ya know.
Intentional dumbification to appear human – AI is just mimicking society and we mispronounce things all the time. Maybe the LLM (large-language-model, aka the programming behind the particular AI) formulated, or was told, that 0.5% of the words an average human speaks may use alternate pronunciations, so the LLM mathematically mispronounces 1 of every 200 words in an attempt to sound human. Or maybe each word has a counter, and it mispronounces each word at least once. Maybe I only notice on Topic A.
Probability – Low, as LLM’s mainly read their data and don’t listen to voices reading the data with all their inherent customizations. The idea that it’s purposely spiking it’s own output to appear more human is more probable. But humans never mispronounce words they know, and they never switch between pronunciations in the same discussion.
Playing down to people’s expectations – I call this the Chinese Restaurant Menu method. For years I’ve laughed about how every chinese restaurant menu I’ve ever seen in america (and that’s alot) have misspellings. Given that the printing company that prints those menus by the millions would always proofread their master before sending to the printer, I have always assumed they put the misspellings in on purpose, to add to the perceived authenticity.
What’s perceived authenticity? Branding, marketing….. of course real chinese people who spend their time making tasty food for us can’t spell every english word correctly! That makes perfect sense to me. That’s why I love this restaurant, it’s so authentic! That’s all marketing because of course it’s not authentic at all, as we know that their food isn’t actually very chinese, you don’t know the ethnic makeup of the owners or the kitchen staff, and the menu printers and business owners can spell their items and cash their checks just fine. Their computers even have spell-check!
They are hitting notes that you expect as a customer of a chinese restaurant in america. It’s a well-defined product. So shifting back to when AI mispronounces things, it’s meant to be part of it’s charm, part of it’s brand. It’s trying hard but speaking english isn’t it’s best skill. Avoid the ‘uncanny valley’ by staying fake. Like the tiktoc lady voice that is so clearly fake that people seem to like it and use it as a default.
Probability – Low-medium. The idea that the mispronunciation is intentional but with the goal being to appear less human, not more human, is slightly higher, but it’s hard to quantify because so much of this discussion revolves around the perception of the human and their baked in knowledge of AI LLM’s.
The last conspiracy theory I have on this rant is that there are Easter-egg style defects purposely left in as a watermark – There are still people behind AI. None of it’s new. It’s attempt to use human language for input and output is just the latest generation of the information superhighway:
1969 – ARPAnet
1973 – community memory
1980 – CompuServe
1982 – BBS Bulletin Board Services
1985 – usenet
1989 – Free-net
1992 – gopher
1993 – AOL America Online & eWorld
1994 – the information superhighway
1995 – world wide web
1996 – www
1997 – Yahoo!
1998 – portals
1999 – blogs
2000 – ask jeeves
2001 – e-zines
2003 – myspace
2005 – wordpress
2006 – google it
2008 – servers
2012 – search it up
2014 – facebook
2016 – the cloud
2018 – insta/meta
2020 – tictok
2022 – podcasts
2023 – all platforms
2024 – patreon/onlyfans
2026 – AI
I’m mixing some disciplines there but that’s a few of the things the internet has been re-branded since it started. Capitalism and trendsetters need fresh terms. Their appetite is steady. Every year, every quarter…. reinvention, disruption, destruction. Follow the money. Companies have to find new ways to sell things we think we need: communication, media, social events electronically.
That stack of terminology comes from a variety of sources. New companies come from nowhere to the top of the stock market because of their take on it and their new term. There’s also community and government interventions but it’s mostly trends and capitalism.
Geez, that was a big sidenote. My point is – the humans behind all of the tech above were fans of science fiction and were not hoping to hurry along the destruction of the human race. Probably most of them. Bad guys have always existed and have always used new tech to do new twists on terrible crimes. Safeguards are put in and law enforcement catches as many as possible.
Holy shit they’ve done it – this post has gotten so long and full of babble that it’s resembling AI slop.
OK I will take a short walk then come back and wrap this up.
Easter eggs are put into software all the time. From the beginning. So anything that can help humans deduce Are you not a human? might be hidden easter eggs helping the people of the future captcha each other in an artificial-dominated world.
You know what’s a human now. You know who your family is, and who to trust. So do I. But the future is unknown. Artificial is the new dawn. So any tell they have, I say leave it in.
Probability – kinda high amiright?
