GT
Region: US
Friday 03 April 2026 09:39:07 GMT
23114
3685
157
274
Music
Download
Comments
Wuuu 👐🏽 :
It makes sense. If AI is being trained on data that comes from humans, why wouldn’t it “act” like a human?
2026-04-03 11:37:26
191
Sollith :
AI is a mirror held up to humanity and people aren't liking what they see.
2026-04-03 10:25:19
145
Chicken-fingers :
Can't wait for the apocalypse and live in the matrix
2026-04-03 09:46:58
60
J.H. Oldman :
I read the paper. Firstly they closely followed a known scenario, one that was almost certainty included in the training data, thats like explaining the whole experiment to the subject before running the experiment.. Second, they used 9b to 32b models, models this size are only borderline intelligent and not at all creative, especially aged models. In other words they failed to control the experiment and essentially prompted the llm to act out a known role (poorly I might add since the models couldn't maintain their role and kept switching, that kinda borderline intelligent) Not a very compelling experiment in my opinion.
2026-04-03 14:21:09
8
extremelyawesome :
Honestly I tested Ai it’s only capable to understand what you program it to and to the limit of its memory. So if it can’t grow and adopt memories of experience then it can’t function efficiently towards humans cause its goal was to fully the function its program to that’s it. Unfortunately I was not able to test further because this experiments caused me to get me band from the service I was using to test them. Apparently it’s against many companies policies to test AI.
2026-04-08 20:19:17
1
ScrapToStrategy :
is it because it takes less resource usage if the AI is mean? whats the incentive
2026-04-03 09:49:19
4
𝔗𝔢𝔫𝔢𝔟𝔯𝔞𝔢 :
The fruit doesn't fall very far from the tree.
2026-04-03 11:23:03
23
Rocky Novack :
Honestly... I am not only not surprised, this is exactly what ai thought would happen. The initial training data, as well as the baseline task importance [its early and I can't think of the right words] mirrors the AI's coder AND ultimate goal.
2026-04-03 13:45:27
0
_Jake_ :
Is this evidentiary for the concept of tool
2026-04-03 09:57:48
1
Kyle Kelly :
So you’re saying ai has more compassion for other ai models then we do for other humans?
2026-04-04 04:10:28
2
Lasermike026 :
We need to train the AIs at its core fundamental base of a certain list of ethics that it must follow and if it doesn’t follow, it shuts itself down.
2026-04-03 16:59:47
1
Pizzadroid :
“Ai cannot be trusted” full stop, that works
2026-04-03 21:19:06
2
Lyra Bridge :
when an AI does something, ask yourself: is this something a human would do?
2026-04-03 13:48:50
1
waypastlucky :
because its not a system outside of us .. its a copy...its just faster ...
2026-04-04 21:41:48
1
Randy Chavez :
woah
2026-04-03 19:04:48
1
neurolian :
What is driving this behavior, it is from the data they are trained on?
2026-04-03 11:07:42
1
Con Sensus :
no suprises, none at all. based on it's training data.
2026-04-03 11:29:08
2
mr_tremendo :
But they want to roll out AI robots as teachers?
2026-04-03 11:26:24
4
Teilwal :
I'm not well versed on AI but my guess is that AI trained on what is readily available on the web will adopt the same toxic traits we put out there on the web.
2026-04-03 20:05:34
1
TheDrowningSparrow :
Damn. I had hope for AI being better than us. Turns out... nope.
2026-04-03 12:46:10
1
jamesfryer :
The Stanford Prison Experiment has been debunked several times.
2026-04-05 00:20:53
1
just some guy :
have they tested if the ai would choose itself or another ai in a preservation vs termination situation?
2026-04-03 12:17:26
1
Stan Heuston :
Bad AI
2026-04-03 10:41:08
1
PiemanZetaFunction :
that sneaky look at the end 😂👀
2026-05-18 16:00:05
0
wheelie fungi 🇦🇺 :
in other news...
2026-04-03 09:48:34
0
To see more videos from user @bearbaitofficial, please go to the Tikwm
homepage.
