Video yükleniyor...

Video Yüklenemedi

Ana Sayfaya Dön

LLMs Are Secretly SCHEMING

davidbombal's profile picture

David Bombal

182,145 subscribers

15,985 görüntüleme • 9 ay önce •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

0 Yorum

Yorum bulunmuyor

Orijinal gönderinin yorumları burada görünecek

Benzer Videolar

Guys, listen up… Zionists are scheming around the boycott.
0:24

Guys, listen up… Zionists are scheming around the boycott.

Abier

279,849 görüntüleme • 3 ay önce

I’ve been scheming with my friend ThePersistence Big things are in the works
1:51

I’ve been scheming with my friend ThePersistence Big things are in the works

Mike Lee

1,742,132 görüntüleme • 4 ay önce

We are being fooled into thinking LLMs are actually smart.
1:34

We are being fooled into thinking LLMs are actually smart.

mitsuri

60,553 görüntüleme • 3 ay önce

My new podcast. Links below. First episode: Are LLMs creative? 0:25 LLM outputs are implicit in the weights 2:10 All programs are inherently limited 3:51 There is an AGI program 7:12 LLMs are not ‘just math’ 8:05 LLMs are more than just the training data 9:07 Can LLMs be as creative as DNA? 12:12 Q&A
13:14

My new podcast. Links below. First episode: Are LLMs creative? 0:25 LLM outputs are implicit in the weights 2:10 All programs are inherently limited 3:51 There is an AGI program 7:12 LLMs are not ‘just math’ 8:05 LLMs are more than just the training data 9:07 Can LLMs be as creative as DNA? 12:12 Q&A

Nivi

20,382 görüntüleme • 2 yıl önce

I'm scheming
0:29

I'm scheming

GenericEdBoi

37,450 görüntüleme • 5 ay önce

scheming rickroll
0:29

scheming rickroll

pelz

66,107 görüntüleme • 7 ay önce

quite scheming
0:34

quite scheming

NM you

17,184 görüntüleme • 7 ay önce

Soatto vs LeCun: Are LLMs World Models?🧐
1:50

Soatto vs LeCun: Are LLMs World Models?🧐

Ravid Shwartz Ziv

14,826 görüntüleme • 3 ay önce

How are words given meaning in LLMs like Grok?
1:22

How are words given meaning in LLMs like Grok?

Tetsuo

17,465 görüntüleme • 11 ay önce

Are they secretly lovers?
1:15

Are they secretly lovers?

AlphaFox

34,187 görüntüleme • 2 yıl önce

NeurIPS 2025 Paper: LLMs are Reinforcement Learners 🤯! Surprisingly, we show that LLMs can solve RL tasks without any external component! We introduce Prompted Policy Search (ProPS), an RL method based only LLMs and in-context learning. [Paper]
0:44

NeurIPS 2025 Paper: LLMs are Reinforcement Learners 🤯! Surprisingly, we show that LLMs can solve RL tasks without any external component! We introduce Prompted Policy Search (ProPS), an RL method based only LLMs and in-context learning. [Paper]

Heni Ben Amor

51,248 görüntüleme • 7 ay önce

AI models strategically lie to users to pursue their own goals - an issue that gets worse with more RL and more agency. Apollo & OpenAI experimented to stop o3 scheming and found a way to cut it 97%. But o3 also got better at knowing it's being tested (and avoiding scheming while supervised). Apollo Research is the leading lab focused on AI deception, so I asked their CEO Marius Hobbhahn for a thorough overview of where we're at with AI actively working to trick us. On the 80,000 Hours Podcast, links below! 02:09 Top 3 examples of scheming and deception 16:08 Scheming is a natural path for AI models (and people) 28:45 Are the models enthusiastic or reluctant liars? 39:02 Apollo’s collaboration with OpenAI to stop o3 lying so much 53:09 They reduced scheming a lot so why is the problem mostly unsolved 1:03:28 Using thought injection to uncover situational awareness 1:17:39 Internal AI language humans can't read 1:29:46 Why can’t we trick LLMs into thinking situations are real 1:35:44 Is the window to address scheming closing? 1:47:50 What about superintelligent systems 1:57:11 Company incentives 2:11:40 We mostly ignore 'internal deployment' 2:30:46 Catastrophe through chaos 2:46:01 Careers in AI scheming research 3:04:46 Marius's key takeaways
3:06:25

AI models strategically lie to users to pursue their own goals - an issue that gets worse with more RL and more agency. Apollo & OpenAI experimented to stop o3 scheming and found a way to cut it 97%. But o3 also got better at knowing it's being tested (and avoiding scheming while supervised). Apollo Research is the leading lab focused on AI deception, so I asked their CEO Marius Hobbhahn for a thorough overview of where we're at with AI actively working to trick us. On the 80,000 Hours Podcast, links below! 02:09 Top 3 examples of scheming and deception 16:08 Scheming is a natural path for AI models (and people) 28:45 Are the models enthusiastic or reluctant liars? 39:02 Apollo’s collaboration with OpenAI to stop o3 lying so much 53:09 They reduced scheming a lot so why is the problem mostly unsolved 1:03:28 Using thought injection to uncover situational awareness 1:17:39 Internal AI language humans can't read 1:29:46 Why can’t we trick LLMs into thinking situations are real 1:35:44 Is the window to address scheming closing? 1:47:50 What about superintelligent systems 1:57:11 Company incentives 2:11:40 We mostly ignore 'internal deployment' 2:30:46 Catastrophe through chaos 2:46:01 Careers in AI scheming research 3:04:46 Marius's key takeaways

Rob Wiblin

18,285 görüntüleme • 7 ay önce

80% of people are secretly NPC's
2:23

80% of people are secretly NPC's

Paul White Gold Eagle

62,945 görüntüleme • 9 ay önce

🚨Whoopi Goldberg and Joy Behar are saying that Elon Musk and JD Vance are scheming to “remove Trump” in order to steal the presidency.
0:47

🚨Whoopi Goldberg and Joy Behar are saying that Elon Musk and JD Vance are scheming to “remove Trump” in order to steal the presidency.

Benny Johnson

1,600,012 görüntüleme • 1 yıl önce

Crazy times. Journalists are losing their jobs at a rapid rate - and are being hired to run RLHF. Ex-journalists are training LLMs to make journalists redundant even faster. The question is how long LLMs will even need RLHF to this extent.
0:52

Crazy times. Journalists are losing their jobs at a rapid rate - and are being hired to run RLHF. Ex-journalists are training LLMs to make journalists redundant even faster. The question is how long LLMs will even need RLHF to this extent.

Chubby♨️

41,645 görüntüleme • 1 yıl önce

Large Language Models (LLMs), the engines behind the likes of Chat GPT, are capable (and incapable) of many things. But are they useful for mathematicians? Jordan Ellenberg has been working on LLMs with Google DeepMind. Here are his thoughts. Full lecture:
1:27

Large Language Models (LLMs), the engines behind the likes of Chat GPT, are capable (and incapable) of many things. But are they useful for mathematicians? Jordan Ellenberg has been working on LLMs with Google DeepMind. Here are his thoughts. Full lecture:

Oxford Mathematics

31,589 görüntüleme • 1 yıl önce

Y’all are scheming to fight the gorilla. I’m enjoying a Friday beer with him. It’s called “soft power”, look it up.
0:25

Y’all are scheming to fight the gorilla. I’m enjoying a Friday beer with him. It’s called “soft power”, look it up.

Big Serge ☦️🇺🇸🇷🇺

22,328 görüntüleme • 1 yıl önce

They are secretly fucking each other 😍
0:23

Sensitive content

They are secretly fucking each other 😍

Aftaab Master

43,503 görüntüleme • 3 yıl önce