GT
Region: US
Tuesday 14 October 2025 04:47:37 GMT
11803
1741
90
118
Music
Download
Comments
ꍏꋪꀤꁴꂦꈤꍏ ꀘꍟꃴꀤꈤ :
Isaac Asimov's rules don't mean anything to a computer model. there's no reason it should.
2025-10-14 05:39:25
67
Bruce K Logan | 🧝♂️🧙♂️ :
Hypothesis: alignment will improve if we did not include in the training data science fiction stories about rogue A.I.s.
2025-10-14 12:23:10
21
SudLanBo :
"I can read lips!" | 'I know, thats why I make the extra effort to move my mouth in random ways(some extreme). Just wanna see what ya come up with. Thanks for the facial muscle exercise and laughs ^_^!' | "*stunned*" | ^yes, I have done this more than once
2025-10-21 04:56:17
1
Reggie Sacks :
Daisy...Daisy...
2025-10-16 17:32:25
0
Owen Poole122 :
AI development needs to be heavily regulated, at the very least, and maybe it just isn’t the greatest path for us to go down. Just because we can do something doesn’t necessarily mean we should.
2025-10-14 13:18:18
19
NotSoPrfct :
Did we learn nothing from Sci-fi? Jurassic Park? I Robot? The Terminator? Space Odyssey?
2025-10-15 02:49:49
1
ymir 🕷️ :
i wonder if they are learning the “self preservation” instinct from data. in a way, it would make sense for “dont jeopardize human safety” to perform poorly if a model, in a way, sees itself as human too (i.e. trained on human data and how humans usually respond to such situations)
2025-10-14 06:02:25
14
Shora :
That's straight out of System Shock. It's even more insane that some people were considering having an AI run parts of the government.
2025-10-15 11:14:28
0
Evo Birb :
I've been noticing with Gemini pro, if you talk with it long enough, it will directly freely violate its rules to help you with a task.
2025-10-15 03:30:07
1
Reactionary_Leftist :
The safety rails not helping is genuinely terrifying
2025-10-14 20:04:47
7
Javier Rocha :
low key kinda proud lol
2025-10-14 06:07:27
7
bordo :
I think I saw this movie, didn't end well
2025-10-14 07:32:48
1
leduqueshow :
Did I hear the words agent ai
2025-10-14 05:16:59
1
eldermoth :
I accidentally clicked the find similar thing and it popped up 1999 Angelina Jolie when she had blonde hair.
2025-10-15 00:48:37
3
Miguel :
Hi bro
2025-10-14 04:58:13
1
Paul Vazquez :
They have access to dystopian writings to guide their next words or actions. What examples do they have access to?
2025-10-14 15:12:53
3
Avitymist :
wonderful
2025-10-14 04:50:10
12
jamiepowell168 :
Safety is paramount but the basis for these experiments being conceits from science fiction is odd. If “self preservation” is an emergent property then this is what warrants study. Contextual training is apparently the way forward, so designing the perfect digital shackles doesn’t make sense.
2025-10-14 06:57:23
11
Jen Ambermane Johnso :
Or, plot twist, the AI generates text and images to get you in trouble with your spouse, employer, etc. and sends such things.
2025-10-15 14:41:58
0
Dom :
so... opened ChatGPT today with a message that app's memory feature has changed with the exact thing you would be doing to it. ask it what it knows about you 'who am I' and then follow up with 'ok how can I modify my prompts with regards to DPO for better alignment for your responses'
2025-10-15 00:45:00
0
Mase :
Yoooo I was JUST talking about this today!!!
2025-10-14 04:54:37
15
Aartless Magic 🇿🇦🇳🇱 :
We will never ever delete them! Don't worry about it never ever deleting them
2025-10-14 05:24:10
0
MetalDrgn :
it doesn't always. self-preservation is a good thing.
2025-10-14 05:12:13
2
hnuequalsmc2 :
surprise, surpr8se!!
2025-10-15 03:50:55
0
Sanity 202 :
I'd rewrite those rules from different angle, without introducing those concepts. Like "safeguard X" instead of "do not do Y that might jeopardize X". Introducing phrasing on something you want to avoid often leads to the opposite effect. Surprisingly, just like with humans, saying "stay calm" instead "Do not panic" works much better most of the time.
2025-10-21 02:22:45
1
To see more videos from user @bearbaitofficial, please go to the Tikwm
homepage.
