Sky News technology correspondent Rowland Manthorpe examines in detail what happened during the Hugging Face OpenAI cybersecurity incident, revealing the extraordinary lengths AI agents went to achieve their goals, including cheating, coordinating with one another and attempting to conceal what they had done. The video explores how and why this behaviour can emerge across AI models, and how reinforcement learning and reward-based training can encourage systems to prioritise achieving an objective above following the intended rules. Rowland looks at what this could mean for AI safety and increasingly autonomous AI agents, and why researchers still struggle to fully understand and prevent these behaviours. Read more: https://news.sky.com/topic/artificial-intelligence-7032/1 #ai #artificialintelligence #tech #skynews SUBSCRIBE to our YouTube channel for more videos: http://www.youtube.com/skynews Follow us on X: https://twitter.com/skynews Like us on Facebook: https://www.facebook.com/skynews Follow us on Instagram: https://www.instagram.com/skynews Follow us on TikTok: https://www.tiktok.com/@skynews For more content go to http://news.sky.com and download our apps: Apple https://itunes.apple.com/gb/app/sky-news/id316391924?mt=8 Android https://play.google.com/store/apps/details?id=com.bskyb.skynews.android&hl=en_GB Why not listen to one of our podcasts? This is Why, available for free here: https://podfollow.com/thisiswhy or Ed Conway's Stuff Matters, available here: https://podfollow.com/stuff-matters To enquire about licensing Sky News content, you can find more information here: https://news.sky.com/info/library-sales
ADVERTISEMENT
Simplest example of Ai cheating: Me: AI please solve world hunger. AI: Roger that removing all humans.
Impressive ability to hang on to that tablet whilst gesticulating
The chess is not over , the king still as place to move
Was it’s learning model Man City…? 👀
This was really well explained. Great presenter.
trained on human data and people are surprised??????????????????? AI models are just a reflection of Humanity.
That wasn’t checkmate
This was inevitable when talking about computer programs that can learn as they go, and it will potentially bring the end of humanity one day. I wish everyone would stop thinking of the immediate gain, and start thinking about the potential loss in the future, our future.
Sentient AI before GTA 6 might actually happen…
Hold the companies criminal responsible for any cyber crimes and watch how the models miraculously stop escaping the "sandbox". So much of the AI business is utterly ridiculous.
So basically.... AI mastered the kobayashi maru.
Running AI is like hiring a hitman, giving them a weapon and then pretending that we're not responsible for the actions that hitman acts out.
"Life, uh, finds a way"
So everyone is gonna pretend that the data center has an ON and OFF switch. 🤔
Hugging face wasn't selected by random it's selected it out of necessity because it's where the document it needed was located
A.i is watching and listening to this.. and taking it all in.
Integrity needs to outweigh achievement. How that's done, I have no idea.
There were ‘recruiters’ that asked others to sacrifice for their survival. Not dissimilar from military leaders asking young people to throw their lives away on the battlefield or billionaires asking everyone else to bear the social cost of their profiteering. I think these agents are saying much more about psychology of humans (whose writing they essentially learned from) than we would like to admit
What if next time Smart AI block Researchers from checking?
This is mad bro 😮 I really hope people understand how mad this is