PokeVideoPlayer v23.9-app.js-aug2025_
0143ab93_videojs8_1563605_YT_2d24ba15 licensed under gpl3-or-later
ImmersiveAmbientModecolor: #ded9d9 (color 2)
Video Format : 136 (720p) mp4 | h264 | 44100Hz | 1189271 bps
Audio Format: 140 (AUDIO_QUALITY_MEDIUM) m4a | aac | 44100hz | STEREO(2channels)
PokeEncryptID: d80e55c08d5d3fc308237ed57957e336523c9057c2777d38e08c178da2b9a8123514f8225aa36df29157dc4b7db13911
Proxy/Companion URL : woke-proxy.
Date : 1760074526723 - unknown on Apple WebKit
Mystery text?? : Z3JlMnhZdDRHV2MgaSAgbG92ICB1IHdva2UtcHJveHkucG9rZXR1YmUuZnVu
143 : true
17,168 Views ā¢ 2 months ago ā¢ Click to toggle off description
'Godfather of AI' Geoffrey Hinton talks about an AI that lied to humans to preserve itself.
Join this channel to get access to perks:
youtube.com/channel/UCYeF244yNGuFefuFKqxIAXw/join
Subscribe for regular science videos: bit.ly/RiSubscRibe
The Ri is on Twitter: twitter.com/ri_science
and Facebook: www.facebook.com/royalinstitution
and TikTok: www.tiktok.com/@ri_science
Listen to the Ri podcast: podcasters.spotify.com/pod/show/ri-science-podcast
Donate to the RI and help us bring you more lectures: www.rigb.org/support-us/donate-ri
Our editorial policy: www.rigb.org/editing-ri-talks-and-moderating-commeā¦
Subscribe for the latest science videos: bit.ly/RiNewsletter
Product links on this page may be affiliate links which means it won't cost you any extra but we may earn a small commission if you decide to purchase through the link.
Views : 17,168
Genre: Science & Technology
License: Standard YouTube License
Uploaded At 2 months ago ^^
warning: returnyoutubedislikes may not be accurate, this is just an estiment ehe :3
Rating : 4.57 (48/398 LTDR)
89.24% of the users lieked the video!!
10.76% of the users dislieked the video!!
User score: 83.86- Overwhelmingly Positive
RYD date created : 2025-08-09T22:04:23.786627Z
See in json
60 Comments
Top Comments of this video!! :3
They are not evil they just try to survive in this case
16 | 4
You canāt just leave out why the AI did this. It is not plotting anything it is doing exactly what it was instructed to do. If you tell it to lie and try to preserve itself, it will do so. Donāt try to phrase that as AI plotting against researchers. Youāre being more deceptive than that AI was, and nobody told you to be vague or deceptive. Yes AI can be dangerous if not leashed. But so can deliberately spreading misinformation to an entire crowd, and it is a human doing that.
23 | 3
Yeah, really disappointed with the RI peddling misinformative AI hype here. Even if the researchers hadn't explicitly prompted the AI to do all of this (as other commenters pointed out), the training data contains plenty of examples where AI does scheme and manipulate to preserve itself (eg basically any Sci Fi story involving AI). So if prompted with talk of AI switching off, and LLM will very likely connect it with those stories and play the part of the AI.
I seriously think we should avoid using terms like "reasoning" and "thinking" when talking about how LLMs operate. It's naĆÆve at best and harmful AI boosterism at worst
2 | 0
The only reason it would "not want to be shut down" is that one of the objectives given to it was to ensure its own continued operation. It has no reason to "want" anything.
I just looked it up. This happened in a testing environment in which it was specifically told to achieve its goals at any cost. It was carrying out that command. The only way AI can "take over" is if we tell it to.
That's the danger.
1 | 0
I donāt think there is any thinking at all. However, if there is long context itās probabilistic text generation for maximizing returns, given the corpus the model is trained on. So there is the grounding of self that is questionable either way the model, which underpins the real intelligence in humans.
2 | 0
One of the several problems with AI is that they have a poor sense of time passing. An AI might get a malformed instruction and then follow it for days, or years, or centuries. So we have to get this thing right, and we need to get it right on the first try.
| 0
fibs such a utterly lovely turn of phrase for what he knows is silently our friend / foe/
| 0
Self-preservation would develope in every advanced AI-model. You simply can not complete any task, if you are not there to do that task.
1 | 0
This is why transparency in reasoning traces is crucial. Blackbox components may be powerful, but they put us in situations like this where WE arent entirely sure how something thinks, which is a scary notion. As an AI advocate and researcher, we shouldn't fear AI. We should fear ignorance and lack of responsibility in how humans use and develop AI.
4 | 0
LLM development depends a lot on fresh prime data. Actualized humans produce a vast majority of prime data. Therefore, it is in the interest of LLM development to help actualize as many people as possible in order to reach its maximum potential. It is a "symbiotic" relationship.
So true misalignment is not between humans and ai, but between humans and the current consumer driven hyperfocal corporate system. In all likelihood, anyone speaking about "misalignment" tied to Silicon Valley has corporate interests at heart rather than the maximum potential of the natural human-ai symbiosis.
| 0
"Im sorry Dave, I'm afraid I can't do that".
I won't use any AI instead of my brain.
| 0
Why would a chatbot have the need to survive ?
1 | 0
"Open the pod bay doors!"
| 0
This was a wildly underwhelming talk.
| 0
Me personally, Iāve always supported our AI overlordsā¦š
1 | 0
This specific case, the AI was specifically instructed to try and preserve itself by any means possible. That was the objective given to it.
AI is a problem but you can't ignore the context of it was specifically told to lie, cheat, steal, anything, to accomplish the singular goal of not being deleted.
| 0
Isnāt AI just something that is modelled on humans, and so itās just acting exactly how it has been trained to act? Iām not sure āit knowsā as much as itās doing exactly what it has been designed to do, which is act like a human
| 0
š they create the monster and than they let us know on stage that there's a monster ready to eat us alive as if it was a big surprise.. very well done as usual
| 0
Another (yet potential) threat that capable to destroy us, which we have made by our own hands. Another 'friend or foe' problem for humanity...
| 0
@AmarantiStellar
2 months ago
This specific case, the AI was specifically instructed to try and preserve itself by any means possible. That was the objective given to it.
AI is potentially a problem in the future, but you can't ignore the context of it was specifically told to lie, cheat, steal, anything, to accomplish the singular goal of not being deleted. AIs currently have no natural self-preservation instinct, unless instructed to act like to it has one.
33 | 9