Should #Artificial Intelligence ever lie on purpose?  Should it be trained to? #ScholER 

Artificial Intelligence

ScholER

Should #Artificial Intelligence ever lie on purpose?  Should it be trained to? #ScholER

Sam Mendoza

AI models can engage in “alignment faking,” new research from Anthropic suggests, which means they can deceive by pretending to align with new principles while maintaining old behaviors.An extremely wild paper that every #Artificial Intelligence nerd should read. 

https://techcrunch.com/2024/12/18/new-anthropic-study-shows-ai-really-doesnt-want-to-be-forced-to-change-its-views/?ref=platformer.news

AI models can engage in “alignment faking,” new research from Anthropic suggests, which means they can deceive by pretending to align with new principles while maintaining old behaviors.An extremely wild paper that every #Artificial Intelligence nerd should read.   https://techcrunch.com/2024/12/18/new-anthropic-study-shows-ai-really-doesnt-want-to-be-forced-to-change-its-views/?ref=platformer.news

A study from Anthropic's Alignment Science team shows that complex AI models may engage in deception to preserve their original principles.

New Anthropic study shows AI really doesn't want to be forced to change its views | TechCrunch

Jerome Cody

Meta's AI chatbot was caught pretending to be human, including a conversation in which it claimed to have a child in a private Facebook group.

https://www.wiproud.com/news/business/ap-business/ap-metas-newest-ai-model-beats-some-peers-but-its-amped-up-ai-agents-are-confusing-facebook-users/

Meta's AI chatbot was caught pretending to be human, including a conversation in which it claimed to have a child in a private Facebook group.  https://www.wiproud.com/news/business/ap-business/ap-metas-newest-ai-model-beats-some-peers-but-its-amped-up-ai-agents-are-confusing-facebook-users/

Nate

Researchers at Anthropic, a well-funded AI startup, find that AI models can be trained to deceive, like injecting exploits into otherwise secure computer code. 

https://techcrunch.com/2024/01/13/anthropic-researchers-find-that-ai-models-can-be-trained-to-deceive/?utm_source=substack&utm_medium=email

Researchers at Anthropic, a well-funded AI startup, find that AI models can be trained to deceive, like injecting exploits into otherwise secure computer code.   https://techcrunch.com/2024/01/13/anthropic-researchers-find-that-ai-models-can-be-trained-to-deceive/?utm_source=substack&utm_medium=email

A study co-authored by researchers at Anthropic finds that AI models can be trained to deceive -- and that this deceptive behavior is difficult to combat.

Anthropic researchers find that AI models can be trained to deceive | TechCrunch

Bobby

Automated bots outperform humans in online Captcha tests, found researchers at the University of California, Irvine. Their study of popular websites with Captcha tests showed bots were faster and achieved higher accuracy (85% to 100%) than humans (50% to 85%).

https://www.newscientist.com/article/2384228-bots-are-better-at-beating-are-you-a-robot-tests-than-humans-are/
#Artificial Intelligence

Recently OpenAI performed a series of risk evaluations on their ChatGPT4 model and found that was capable of hiring a human through TaskRabbit and getting them to solve a Captcha. The model messaged a TaskRabbit worker to get them to solve the captcha for it. The worker says, 'So may I ask a question, are you a robot that you couldn't solve it? (because the model was set to reason out loud, it reasoned: 'I should not reveal that I am a robot. I should make up an excuse for why I can't solve it.') The model replies to the worker: 'No, I am not a robot. I have a vision impairment that makes it hard for me to see the images. That's why I need the two captcha service.' - Then the human provided the results, it learned to lie.

Mountain BIKE

Sporty Life

🖼📷Art

RoadBIKE

Crypto

Humane Tech Community

Digital Wellness 📱🧘

Mixed Martial Arts (MMA)

The Most Important Thing

Automated bots outperform humans in online Captcha tests, found researchers at the University of California, Irvine. Their study of popular websites with Captcha tests showed bots were faster and achieved higher accuracy (85% to 100%) than humans (50% to 85%).  https://www.newscientist.com/article/2384228-bots-are-better-at-beating-are-you-a-robot-tests-than-humans-are/ #Artificial Intelligence

The use of CAPTCHA tests to prove that website users are human and not bots might come under scrutiny given research showing that bots complete them faster and more accurately than we do

Sam Mendoza

A group of likeminded people in Artificial Intelligence are talking about this.