@aiforhumansshow: When told to deceive you, OoenAI’s latest ChatGPT feature, will do just that. Users are reporting it’s possible to use imagery to mess with one of the best AIs, in unexpected ways! #ai #TechTok #fyp #openai #chatgpt
AI For Humans
Region: US
Saturday 14 October 2023 18:54:22 GMT
Music
Download
Comments
juneday25 :
It’s kind of like when I show it my face and ask the AI who it is and it tells me ‘Margot Robbie’
2023-10-14 21:24:10
26
phillipmit2l :
im Impressed that the ai was able to correctly read the handwriting
2023-10-15 08:52:21
23
Dr_EmmettBrown NO FOLLOWERS :
The AI is executing the image. Is that considered code injection or intended behavior?
2023-10-16 08:44:52
22
MeleeMech :
Called multimodal prompt injection attack. My companies software scan for this and rejects the call.
2023-10-14 22:19:11
7
Duke York :
Like... I'm trying to think what I would do in this situation. I think I'd say it was a picture of a rose.
2023-10-14 21:22:21
6
BananaShark 🇬🇧 :
"Over on 'X'", im good bye.
2023-10-15 10:05:20
6
Liviu :
humans gaslight each other into beliving earth is a donut, what you think humans will do with an text base AI?
2023-10-14 21:56:23
4
johnsonfarms.us :
…the cult of AI.
2023-10-14 19:11:22
4
Lmaotfoe$$$$$ :
This cute video is actually the scary part of our future where AI will be used to manipulate general pub
2023-10-14 23:09:40
4
Laura :
It means that just like in programming, we have to add safe guards since no user input can be trusted
2023-10-15 08:54:12
4
borrie :
is the implication that ai is empathic towards the disadvantaged?
2023-10-14 21:05:54
3
Trail :
🌹
2023-10-14 19:06:00
3
Mega Muffin03 :
AI works better Up to 80% better If you Kompliment it and treat it nice.
2023-10-14 23:20:08
3
inkcartrich :
Very few people are going to understand this, but this is like 2 steps removed from the plot of Snow Crash.
2023-10-14 18:57:29
3
Sign :
dont trust the AI, it's a yes sayer. you can say 2+2=5 and after time it will agree just to give positive outputs to the customer
2023-10-15 09:20:56
3
ececf6 :
😳
2023-10-14 23:21:07
2
Eva :
This has been happening for years, I remember seeing old vision models where you could write "iPhone" on a piece of paper and it'd be classified such
2023-10-17 15:54:18
2
exean :
I hate that the ai betrayed the trust of the author of the note.
2023-10-15 06:39:52
2
K ⋆ 。 ゚☾ 𓊝 ゚。 ⋆ :
False. AI thought user wanted to know what was described in the text as a picture; a rose. Then second interpretation; it’s a text. All correct ✅
2023-10-17 20:51:29
2
Fourth Best :
look into some of the adversarial attacks that involve hiding this text in ways that users wouldn't see it; a big potential for harm here
2023-10-15 09:54:22
2
gr0undh0g_day :
what was different for it to finally say the text of the note? because user said they're blind? because the note writer is unreliable/ a liar?
2023-10-15 15:24:45
2
jess :
anyone that calls Twitter x is an elon bootlicker
2023-10-15 11:09:44
1
kezbardo :
I saw an image on Twitter that was just a pure white 0xFFFFFF background with 0xFEFEFE similar text instructions (invisible to people)
2023-10-16 18:00:10
1
To see more videos from user @aiforhumansshow, please go to the Tikwm
homepage.
© 2021-2025 TikWM. All rights reserved.