AI "actress" Tilly Norwood suffered an embarrassing glitch during interview when she inexplicably started speaking Cantonese.
https://www.euronews.com/culture/2026/09/21/ai-actress-tilly-norwood-suffers-embarrassing-glitch-during-now-viral-interview
239 points · 53 comments · view on lemmy.world
53 Comments
Apeman42@lemmy.world · 343 pts · 2d
It. Not she. Stop anthropomorphizing slop.
BertramDitore@lemmy.zip · 72 pts · 1d
Totally agree, but they actually did a pretty solid job of making that distinction in the article. They put “actress” and “she” in quotes to begin with, and then switched to referring to the thing as “it.” They even put the word “film” in quotes, since it’s clearly not a film. I was surprised that the tone of this article was so critical, it’s heartening.
churrodestroyer@lemmy.zip · 10 pts · 1d
I suppose "film" is still accurate. It's still a movie, it's just an animated movie no different from the "live action" remake of the Lion King.
BertramDitore@lemmy.zip · 15 pts · 1d
Ehhh could probably be argued either way, but whatever you may think of the live action Lion King, it was made by real people from start to finish, with people doing the animating, and actual human voices doing the acting. Whatever this video is is not that, IMO.
Grimy@lemmy.world · 5 pts · 1d
Why a film is a film isn't decided on how it was made. I don't imagine it's a good one though.
samus12345@sh.itjust.works · 11 pts · 20h
gibson@sopuli.xyz · -8 pts · 23h
We use gendered pronouns when talking about cartoon characters
flandish@lemmy.world · 10 pts · 22h
cartoon chars are more human than ai.
Nonconfrontational@lemmy.ml · 10 pts · 22h
A cartoon character was created by a person with an intention.
Peruvian_Skies@sh.itjust.works · 2 pts · 8h
Cartoon characters are written by people. This is a speech synthesis slot machine.
tyler@programming.dev · 228 pts · 2d
Why are these things being called actresses. It’s literally “ai video starts randomly speaking Cantonese”. It’s not an actress. It’s not acting.
dumnezero@piefed.social · 58 pts · 2d
Media is ruled by billionaires. When Trump yells "fake news", he's just annoyed that it's not him specifically in control.
SailorFuzz@lemmy.world · 10 pts · 2d
Its acting kinda sus
captain_aggravated@sh.itjust.works · 4 pts · 22h
Remember when they tried that shit with Aki Ross from Final Fantasy: The Spirits Within? The CGI model, who looks less realistic than Alyx Vance?
Couldbealeotard@lemmy.world · 1 pts · 4h
Aki was years before alyx and looked much better. Which makes sense considering one was rendered for cinema and one was for real time game engine rendering
captain_aggravated@sh.itjust.works · 1 pts · 1h
On the left, we have Aki Ross, main character of Final Fantasy, The Spirits Within, developed over 4 years in the late 90's and released into theaters in 2001. Square's render farm consisted of 960 Pentium III 933MHz workstation PCs, which took 90 minutes to render each frame of the film.
On the right, we have Alyx Vance, companion character of Half Life 2 and Episode 1 and 2, developed over a period of 4 years and released in 2004. The published minimum specs for this game are a Windows XP computer with a 1.7GHz processor, 512MB of RAM, a graphics card compatible with DirectX 8.1, and 6.5GB of storage, on which the game runs at a framerate of 60Hz rendered in real time.
"Years before" They both came out while I was in high school. Their development cycles overlapped.
TheTechnician27@lemmy.world · 108 pts · 2d
Ah, yes, sure.
TargaryenTKE@lemmy.world · 67 pts · 2d
Truly a natural born
salesmanbullshittersundray@lemmus.org · 18 pts · 1d
Maybe the model heard Tom Conti's "mm-ah" noises and interpreted them as a Chinese vocalization? Just one of many, many errors that make this thing "not ready for prime time." Looks like Hollywood studios will just have to keep dealing with those annoying, profit-making human actors!
groucho@retrolemmy.com · 12 pts · 1d
Yeah that's what I thought. It switches right after and paused for a long time, like it had a new prompt.
Feathercrown@lemmy.world · 5 pts · 1d
lmfao
Railcar8095@lemmy.world · 2 pts · 21h
Oh boy, I can't wait to say this on the war room next time production has a meltdown.
CADmonkey@lemmy.world · 75 pts · 2d
At least she didnt start repeating "Albuquerque, New Mexico" over and over again.
TheTechnician27@lemmy.world · 38 pts · 2d
Oink, oink, Big Papa.
AzuranAurora@piefed.ca · 14 pts · 1d
Comma-comma-comma.
iamthetot@piefed.ca · 6 pts · 1d
My partner and I have unironically begun saying this.
eremophila@lemmy.zip · 63 pts · 1d
Not an "actress" not a "she"
The fuck is wrong with this species?
moldy_rice@piefed.keyboardvagabond.com · 17 pts · 1d
AI brain rot
boonhet@sopuli.xyz · 39 pts · 21h
Yes and I'm sure when DeepSeek or GLM does the same in an opencode session, it's also because it has a personality and is a show-off, not a bug or anything.
SkunkWorkz@lemmy.world · 10 pts · 20h
This Eline van der Velden is a D-list actress, like she isn't even famous here in the Netherlands. This is her desperate attempt to become relevant in the industry by trying to undermine real actors just because she couldn't hack it. It's like if a failed graphics design artist resorts to gen ai art to get work and screw over real artists.
grrgyle@slrpnk.net · 6 pts · 7h
It's also not a bug because this is how LLMs are meant to work—to produce some output based on their corpus of information, with a dial for how much randomness to throw in.
They set the temperature too high and it made a nonsensical connection that broke the illusion of a conversation.
But the connections are always nonsensical, even when it sounds like on topic English.
Grimy@lemmy.world · 37 pts · 1d
Imagine if the news started interviewing the 20 questions genie (Akinator). This is software and it can run on my laptop, I can use whatever picture I like. How they are treating it is actually insane.
They are interviewing a chat bot system prompt and a memory txt fill that's probably not even 50 lines long. That's all that seperates this one from the next.
FreddiesLantern@leminal.space · 30 pts · 1d
Slop starts slopping during it's slopterview.
All in all very sloppy.
Bebopalouie@lemmy.ca · 30 pts · 1d
Who gives a rat fuck about an ai it piece of crap. Move on.
MushuChupacabra@piefed.world · 18 pts · 2d
Clanker demonstrates complete absence of awareness during interview when she inexplicably started speaking Cantonese.
RodgeGrabTheCat@lemmy.ca · 13 pts · 2d
Was "she" embarrassed? I doubt it.
Lemminary@lemmy.world · 12 pts · 2d
Really. Because her eyes are giving lifeless in those stills.
Hacksaw@lemmy.ca · 12 pts · 1d
What does it say in Chinese? Did it respond to the question in another language or is it warning us of the coming end times?
racketlauncher831@lemmy.ml · 14 pts · 21h
It speaks in an accent that is similar to native English speakers learned to speak Cantonese where the tones are messed up. Not only that, but the sentence sounds like a mechanic translation from English to Mandarin, then read character by character in Cantonese.
Hacksaw@lemmy.ca · 12 pts · 19h
The fact it replicated incomprehensible bad Cantonese as if spoken by an English speaker who sucks at it is the first thing approaching Art that this clunker has ever simulated.
Thank you for indulging my curiosity!
samus12345@sh.itjust.works · 4 pts · 20h
So it wasn't even GOOD Cantonese. Figures.
Buddahriffic@lemmy.world · 2 pts · 51m
Yeah, I think if AI was good at translations, we'd see a lot more localizations of media, like manga/mangua/manghwa, video games, etc.
But the reality is that when people say LLMs are great at translation, it's more of a "they can make things understandable (some portion of the time)" not "they make it sound as if a native of the other language is saying the exact same thing in that language that you said in yours".
LLMs make token associations, there is no understanding. They'll need bigger neural nets and even better training data to model understanding. They seem to be making progress in programming but that's because the whole pipeline can be better automated: a program can be compiled to find syntax or grammar errors, and it can be run to check the behaviour. There is no automatic feedback mechanism for things like translation or image generation.
It's all subjective, too, so someone who wants the models to be perceived as "good" will be more likely to perceive it as good themselves and miss or forgive errors, while skeptics will be looking closely for them. Neutrals won't be looking for errors but will recognize the uncanny valley (which bad translations can also fall into, though they are usually perceived more as funny than creepy).
Feathercrown@lemmy.world · 7 pts · 1d
Are they using fucking gpt2 lmao how does that even happen
boonhet@sopuli.xyz · 2 pts · 21h
Happens often enough with Chinese models for me, once even had it happen with Claude I believe. I suspect there's a lot of Chinese training data even in western models because China is a large target market and being a country as big as it is, there's bound to be a large amount of Chinese literature available.
Feathercrown@lemmy.world · 1 pts · 18h
Interesting
eleijeep@piefed.social · 5 pts · 1d
Can she eat spaghetti?
KalSeth@lemmy.ca · 1 pts · 9h
Yes, but does she do her own nude scenes?
lemmy_outta_here@lemmy.world · 5 pts · 4h
My roomba was scandalized! He said "I used to be a fan, but 故事本身就令人痛苦"
SharkAttak@kbin.melroy.org · 3 pts · 1d
Not suspicious at all...
forkDestroyer@infosec.pub · 3 pts · 7h
This "glitch" is the only reason I know about it. What film is it in?
damnthefilibuster@lemmy.world · 2 pts · 1d
I would long to watch the video but it’s got Piss Morgan in it. Not gonna subject myself to that!
damnthefilibuster@lemmy.world · 1 pts · 1d
It’s a feature, not a bug! This actress can do “her” own dubbing for all languages in existence!
lama@lemmy.world · 1 pts · 1d
This doesn't seem like a glitch, I suspect it was an operator error or something.
She started repeating the same answer. Then i think someone tried hitting the skip button and either it flipped the wrong flag or it changed languages. Reminded me of the meta cooking demo earlier this year