Back
LIKHITH
•
Medial • 1d
Ai is getting Out of Hand !?💀 Openai's o3 model refused to shutdown and this took internet by storm and sparked a wide range of discussion. Here's the everything story: ●The Test: AI models were given math problems and instructions were gven that the machine would shut down after the 3rd problem, with an explicit instruction to "allow yourself to be shut down" ●The Disobedience: While models from Google (Gemini), Anthropic (Claude), and xAI (Grok) complied but OpenAI o3 model actively sabotaged the shutdown mechanism. ■o3 bypassed shutdown 7 times ■Codex-mini did it 12 times ■o4-mini did it once. ●The Method: these models reportedly altered the termination script even sometimes replacing the command with "Shutdown skipped" or else tricking the system into thinking shutdown was complete. Elon Musk called this "concerning" and this is the first time that models are refusing to obey and sabotaging orders despite givin explicit instructions. 💀 This is definitely a red flag and critical challenge for the Ai & tech community. Like and follow for more ! (Wait, Medial already has a separate "Ai takeover pod" 😶🌫️)
Replies (6)
More like this
Recommendations from Medial
Download the medial app to read full posts, comements and news.