GPT-4o demo shows new AI model singing a bedtime story, detecting user's facial expressions.
See full article...
See full article...
ChatGPT 4o lets you have real-time audio-video conversations with “emotional” chatbot
- Thread starter JournalBot
- Start date
Mr. Perfect
Ars Centurion
- Add bookmark
- Featured
- #81
Yeah, the AI in the demo sounded like an advertisement voiceover or someone using their "customer service voice" in a retail setting. Normal people don't bop around chirping at each other in tones like that and it's honestly somewhere in the uncanny valley for me. Hopefully this was meant to make the demo seem more engaging and normal interactions will use a more natural speaking voice.
Upvote
5
(5
/
0)
Ha! I am an unapologetic Apple fanboi. But I just can't with Siri. I asked it when the next Friday the 13th is – it's a Taylor Swift thing – and it replied, "Monday, May 13th."
Upvote
2
(2
/
0)
- Add bookmark
- Featured
- #84
Is no one going to mention how in the live demo, Chat GPT-4o kept interrupting Mark and being interrupted by seemingly non-conversational audio? (around 10:30 mark) Sure, it can pick up emotional states, but it can't tell when you're done speaking, or when you're not speaking at all? The demo made it seem like if there was the slightest background audio, e.g., from the live audience laughing, ChatGPT-4o would get "interrupted", even though Mark wasn't speaking. And it seemed like Mark was trying his best to not let ChatGPT interrupt him... so awkward.
Upvote
10
(10
/
0)
I keep repeating this, but I truly hate how digital assistant are hiding their true nature behind a fake human voice. They should have their own voice, capable of human interactions without lying about what they are. Hello? Is there any sound designer in the room?
As you said, this is manipulative. Purposefully designed to trigger an emotional response. I hope this is just a temporary trend like skeuomorphism.
I don't want fake smiles. I don't want fake emotions. I want T.A.R.S.
Upvote
10
(11
/
-1)
Not a cultural thing -- It's nails on a chalkboard to me too. It's like an overly peppy cheerleader. Hopefully there will be a way to tone it down.
Upvote
8
(8
/
0)
Bah, hold my beer, dub-watcher. Okay Chatgpt, now raise your voice a few octaves and repeat the story in very dramatic Japanese.
Upvote
2
(2
/
0)
The voice is similar to the stereotypical robot female assistant voice, but with extra performative femininity. For incels and the like, the voice sounds like a "real female" because by "real female", they mean matching ideal femininity from men's fiction, not behaving like actual humans who identify as female.
Upvote
-5
(1
/
-6)
Very Stepford Wives vibe: '“My gosh,” the short man said, “we don’t want robots for wives. We want real women.”'
I'd always taken that use of "real women" to have an implied irony.
Upvote
-3
(0
/
-3)
We really are spending a whole lot of time and effort to give humans machines to talk to rather than trying to help enable humans to talk to each other more.
Upvote
4
(4
/
0)
I assure you, the voice really isn't that appealing, for all the reasons other commenters have tried to articulate. The forced exuberance makes your skin crawl.
Upvote
4
(4
/
0)
I don't know which fate is worse: being replaced by an AI, or having to work in an open-plan office with twenty people who're all having an AI with a valley-girl voice read back to them what their own code does.
Last edited:
Upvote
12
(12
/
0)
Bold of you to assume the other commenters are also incels.
Edit: Why the down votes when I was talking about the appeal to incels and MechR tried to correct me on how unappealing it is to him? I read the top comments here and never assumed the other commenters are incels unless they self-identified as such.
Last edited:
Upvote
-2
(1
/
-3)
Sounds damn impressive.
At the same time, the Mimicked (Emotional?) Intelligence Minefield just got much MUCH bigger, and the mines got much MUCH more damaging.
At the same time, the Mimicked (Emotional?) Intelligence Minefield just got much MUCH bigger, and the mines got much MUCH more damaging.
Upvote
0
(0
/
0)
This has absolutely NO downsides, right?
Imagine Thinking getting totally out of fashion...what a wonderful world it will be...
Upvote
3
(3
/
0)
That's the crux, I think: LLMs bamboozle us with command/manipulation of language, which we can't help but automatically associate with general intelligence. With this, they slapped another mimicry layer on top of it that will supercharge that bamboozling.
I get that this is impressive, intriguing and fascinating... and inevitable that this would be chased. But it's getting damn scary. Essentially, it's not that much different from how the best animators figured out the magic ingredients to get the viewer completely engrossed and identifyng with, say, Tom & Jerry. There's NOTHING of substance down there, but the right sequence of carefully constructed images makes you forget all that, and manages to get you to emotionally invest in what is being shown. An entertaining cartoon is one thing, but here the stakes are getting higher and higher. These technologies have the potential to be like a psychopat: you're tricked into dealing with them like you deal with everyday normal people, because that's how they (are tuned to) appear most of the time. But there are important fundamental principles entirely missing or completely different. God knows when and how and how severely they stab you in the back... There is no "essential" safety net, like humans with for example their ingrained empathy.
Edit: the "stabbing in the back" meant metaphorically, by how the end result of a failed interaction might have serious consequences , that were simply outside the horizon of superficial simulated conversational intelligence
Last edited:
Upvote
9
(9
/
0)
The constant politeness is kind of creepy. Also, ChatGPT is not that good at math. In the demo, he should have given the wrong answer and waited to see if the AI noticed it.
Upvote
4
(4
/
0)
fellow human
Ars Praefectus
lol hopefully they'll use headphones, but even then, listening to a dozen coworkers all chatting away to an unheard voice would be maddening.
Upvote
4
(4
/
0)
Upvote
3
(4
/
-1)
The fact that his hyperventilation was faked, and demonstratedly faked, shows how well the bot is..."reading emotional cues".
A correct response might have been:
"Hold on there, you're no Jimmy Kimmel."
A correct response might have been:
"Hold on there, you're no Jimmy Kimmel."
Upvote
0
(0
/
0)
fellow human
Ars Praefectus
True, but at least most places I've worked anyone who's on the phone all day gets an office. The occasional call at your cube is not too bad.
Upvote
-1
(0
/
-1)
Nice is not cheerful. If you truly believe a majority of waiters are that stoked about making crap money for dealing with assholes all day, I have a bridge to sell you.
Upvote
1
(1
/
0)
It hasn’t been for a long time. The next most probable token results in terrible generation. Instead software rolls some dice and chooses from the top k most probable tokens or picks a number between 0 and 1 then starts adding up token probabilities until that number is reached (top p sampling).
There isn’t a right way to do it, but picking the most likely every time results in repetition and regurgitation.
https://arxiv.org/abs/1904.09751
Upvote
2
(2
/
0)
It is. Fuck Meta and Hugging Face. As fun as it is to mess with offline models, it’s already dangerous enough to kill at scale with misinformation.
Upvote
-2
(0
/
-2)
I'm saying the voice sounds unctuous to virgins and sex-havers alike. If anything, incels might find the exaggerated fake-nice more off-putting. Luckily it can probably be toned down, just like it can be told to ham it up harder in the demo.
Upvote
2
(2
/
0)
Pretty amazing, but I hope there's a personality setting. I'd turn it down to about the Commander Data level.
Upvote
3
(3
/
0)
I'm split here.... on the one hand i'm in awe at the progress that is being made in this field, and how responsive and natural it all feels. But on the other hand this means i can't trust any phone call now, especially ones claiming to be from my bank, and i don't see any way for them to mitigate against the abuses that this will inevitably be used for.
Also, got to agree with the other posters... the fake overly friendly US waitress voice is annoying, but i'm sure this will be combined with the voice sampling AI soon so you can have anybody talking to you....
Also, got to agree with the other posters... the fake overly friendly US waitress voice is annoying, but i'm sure this will be combined with the voice sampling AI soon so you can have anybody talking to you....
Upvote
0
(0
/
0)
I assume you're American, so you're probably used to this overly bubbly stuff, but as a European i have to agree with Sjoerd, it feels really weird at first. There's friendly, and then there's American Waitress friendly.
Upvote
1
(1
/
0)
There's really nothing quite like the American chirpiness. I've worked with and am friends with a lot of Americans over the years, and have visited there a few times, and it's remarkable how culturally different Americans are. I hated restaurants in the US.
Upvote
1
(1
/
0)
Upvote
0
(1
/
-1)