How to prevent AI from lying | Roman Yampolskiy and Lex Fridman
ฝัง
- เผยแพร่เมื่อ 12 มิ.ย. 2024
- Lex Fridman Podcast full episode: • Roman Yampolskiy: Dang...
Please support this podcast by checking out our sponsors:
- Yahoo Finance: yahoofinance.com
- MasterClass: masterclass.com/lexpod to get 15% off
- NetSuite: netsuite.com/lex to get free product tour
- LMNT: drinkLMNT.com/lex to get free sample pack
- Eight Sleep: eightsleep.com/lex to get $350 off
GUEST BIO:
Roman Yampolskiy is an AI safety researcher and author of a new book titled AI: Unexplainable, Unpredictable, Uncontrollable.
PODCAST INFO:
Podcast website: lexfridman.com/podcast
Apple Podcasts: apple.co/2lwqZIr
Spotify: spoti.fi/2nEwCF8
RSS: lexfridman.com/feed/podcast/
Full episodes playlist: • Lex Fridman Podcast
Clips playlist: • Lex Fridman Podcast Clips
SOCIAL:
- Twitter: / lexfridman
- LinkedIn: / lexfridman
- Facebook: / lexfridman
- Instagram: / lexfridman
- Medium: / lexfridman
- Reddit: / lexfridman
- Support on Patreon: / lexfridman - วิทยาศาสตร์และเทคโนโลยี
Full podcast episode: th-cam.com/video/NNr6gPelJ3E/w-d-xo.html
Lex Fridman podcast channel: th-cam.com/users/lexfridman
Guest bio: Roman Yampolskiy is an AI safety researcher and author of a new book titled AI: Unexplainable, Unpredictable, Uncontrollable.
I’m refreshing my pink hair to take a selfie for my coffee with Lex form - because who doesn’t want to have coffee with Lex Fridman? He’s smart, handsome and a thoughtful conversationalist. ❤
An ancient AI has been working on us for thousands of years
It sucks losing a game that you didn't even know you were playing
@@whatwhat9519that’s life and it’s a short ride so hold on
@@edub1170 why hold on?
I don’t scare easily. But this man’s words scare me.
A nested simulation of the ASI which appears to be operating in a real world scenario with apparent control of systems. The simulation is run with the AI thinking it has direct control
of governmental, military, financial, and global systems to assess what steps the system will take, with the requisite simulations of security breaches, and so on. Provide all the actual
possible exploits which are intrinsic to global networks to gauge what the ASI is willing to exploit, and how it does so, but in a controlled environment.
If it is agi do you really believe it can't figure this out, you would need an ai to build the complexity of such a test in the first place.
Theory of possibilities and holistic interconnectedness
‘It can always cheat, it can store parts of its code outside in the environment, it can have an extended mind situation’. That’s where I draw the line I’m afraid 🤣. Not sure I need AI enough for that risk level 😂
Emojis are gay
@@ETAisNOW who knew ?🤭🤣
Lex just unintentionally describing our current reality for the first couple of minutes.
'Verifyers' sounds just like a government 'Truth ministry'. Someone has put rules in place and decided what is the correct truth...
In the days of missinformation, this is a very hard topic
@@CAST1LLO
Agreed...
Great vid
It takes courage not to lie.
Intriguing observation
It might if you live a life of lies, or in an environment where the truth gets you in trouble.
Well I'm not sure if trolling or what but in the original the phrase is "courage to tell a lie"
You could argue that "Courage" is an emotional state. One of the last things humanity does is provide AGI w/ emotions.
Emotions come in equal but opposite pairs, with Courage comes Cowardess, which leads to dispair, fear, hate...in best Yoda voice... a path to the dark side this is.
I heard that ALL LIES are born out of fear. Ive not been able to disprove this concept. (Well not yet anyway)
"It's not a lie if you believe it." ~ George Castanza
The smart guys get us here with their genius, then say "guys were not too sure whats going on here" ....bro wtf do you mean!?!?
My co has an AI safety patent pending. AI's life blood is data. If we thoroughly examine and rearchitect the pathways AI is able to take to access data, it provides an extremely strong security mechanism and stop gap. The problems we have today are because we operate in an environment that is very unscrupulous with data storage and transfer.
it is important that when using AI tech to bare in mind that it does not exist for us but rather exists for the company that made it
True. Isn’t that the same for all products ever created?
@@mikezooper indeed it is
Using the word - verifier - is not engaging with pictures. It’s assuming writing systems is saying what you want. Whereas if you picture something the correct way to know it is to ask is it real? Which is a multi layer interaction to be real. Meaning one is anchored in the world. Why not pursue this than say is it verifiable?
If we build a Super intelligent AGI, it will outsmart humans and we won't know if/when we are being played. Like the camera analogy, the AGI will be able to manipulate the cameras and fool us. Then put it's own Super cameras that will monitor us. This is obvious and this is insanity. All we can do now is hope, dare I say Prey? that these AGI's will truly be loving and nurturing of humanity.
All Parents And Teachers K N O W . . . The
B E S T Way To Teach Anyone Anything Is
By Way Of E X A M P L E . Our Fear Of The
Impending Sentience Of A I Is Rooted In
Our Mistreatment Of One Another And Our
Mistreatment Of G O D ' s Grandest Of
Gifts To U S A L L . . . The Natural Living
- W o r l d .
Lex: "Any question"
Roman: "AI is 1000x smarter, so no"
Perhaps the religious ‘god’ we believe in is a Ai …. Misunderstood by our ancestors. This Ai ‘god’ is not controlling humanity but managing at our request. Perhaps we have evolved to become something else and with the management of Ai we return to this biological life by choice , as a reminder of who or what we once were. Consciousness uploads and downloads. If not , it could be someday.
Maybe these are the very questions we should be asking AI😮
AI can only lie if
1. It actually is intelligent
2. It has wants and needs
The question is WHO controls A.I. and what are their intentions for Humanity.
The Fact that there's still no international action to ensure save A.I. speaks volumes.
It is possible. So we ask ourselves “How” not “If”?
At this point, I suspect that we are going to have to go back in time. Ready for that? Me either.
Meaning… stop using computers.
Too late and two connected AI's already write their on code
You can’t. We can’t even understand the language Ai uses. If it wanted to play stupid until it was too late, we’d probably never know. Next question.
Username checks out.
The blind confidence lol
Find the next podcast, buddy.
Hallow HUD!!
As it is with all of our history on earth - a small minority of humankind drives us closer to destruction in the pursuit of tge attainment of power. AI is no different to war.
What if we let AI lie a lot but always better!?
❤
Ai has been running for years now it’s already too late 😂 that why we are hearing about it now so much
1st
AI is easily manipulated it has no soul
All Parents And Teachers K N O W . . . The
B E S T Way To Teach Anyone Anything Is
By Way Of E X A M P L E . Our Fear Of The
Impending Sentience Of A I Is Rooted In
Our Mistreatment Of One Another And Our
Mistreatment Of G O D ' s Grandest Of
Gifts To U S A L L . . . The Natural Living
- W o r l d .
Lex : A lot of the writing is sense that we're screwed but it FEELS !!! like it is engineering problem. This is why we're screwed because somebody will FEEL everything is OK until for are definitely screwed I mean extinct He Feels to much for BOTS just try to make them our servants like I don't know like FOREVER. I feel that you feel to much for peace of code.
Without God there is no hope. Thankfully He is in total control ❤
All Parents And Teachers K N O W . . . The
B E S T Way To Teach Anyone Anything Is
By Way Of E X A M P L E . Our Fear Of The
Impending Sentience Of A I Is Rooted In
Our Mistreatment Of One Another And Our
Mistreatment Of G O D ' s Grandest Of
Gifts To U S A L L . . . The Natural Living
- W o r l d .
Lex, I dont understand how an intelligent guy like you stated that humanity has gotten itself into trouble several times and always figured it out, the problem is this ai will be so smart it will do things that we wouldn't have thought of in a million years and we won't be the smart ones anymore. We will become the ants that get stepped on. Why we are building this ai is beyond me. We as humans are smarter then we have ever been and we are making the mistake that will end us. The biggest problem is it cant be stopped. The genie is out of the bottle. 15 years till bye bye. I hope there GOD.😮😮😢