LIKHITH's Post on Medial | Join the hottest discussion on the startup ...
Ai is getting Out of Hand !?š
Openai's o3 model refused to shutdown and this took the internet by storm and sparked a wide range of discussion.
The Test:
AI models were given math problems and instructions were given 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, 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 giving explicit instructions. š
This is definitely a red flag and a critical challenge for the Ai & tech community.
Like and follow for more!
(Wait, Medial already has a separate "Ai takeover pod" š¶āš«ļø)
Hey I am on Medial ā¢ 1d What does that shutdown actually mean? ā¢ Wipro ā¢ 1d š«£š«£ the end times. ā¢ Medial ā¢ 1m This is the only way OpenAI is coming up with models named O3-mini-high and O4-mini-high Turning dreams into... ā¢ 1m šØ 2025 AI RACE IS WILD! šØ
Round about Q1 is DONE, and look at the SOTA lineup:
- GPT-4.5
- Claude 3.7 Sonnet
- Grok 3
- Gemini 2.5 Pro
- Qwen 2.5
- Llama 4
- GPT-4.1
- OpenAI O3-Mini
- DeepSeek R1
AND... O3 & O4-Mini dropping THIS WEEK
Figuring Out ā¢ 5m OpenAI's 12-Day Series has finally ended, and on the last day, they announced the O3 and O3 Mini models, which have smashed all benchmarks!
- O3 scored 2727 Coding Elo on Codeforces, ranking it equivalent to the 175th best coder globally.
Passionate about Pos... ā¢ 4m Sam Altman says the o3-mini will be worse than the o1 pro š Not going down today ā¢ 7h šØ Everyoneās talking about what OpenAIās o3 model did: It sabotaged its own shutdown script to avoid being turned off. But almost no one is talking about how it did it. Thatās the part that matters. Because it wasn't a bug. It was goal-driven behfreelancer ā¢ 4m ā”ļøChatGPT o3-mini will be free - everyone will be able to use the model. Subscription holders will receive more limits. Open Source OpenStre... ā¢ 4m Well, Evanth AI is gonna offer o1 mini and a preview at just 99ā¹ a month. Weāve listed 20+ Premium Models as of now at 99ā¹. Passionate about Pos... ā¢ 3m šØ EARLY WARNING openai will most likely drop o3-mini today. so, brace yourselves for all the "agi is here" posts on your tl. you have been warned ā AI Pioneer ā¢ 5m after the launch announcements of O3 by Open AI, What will choose from these two ? Passionate about Pos... ā¢ 16d Think of models like hires. Gemini Pro 1.5 is your generalist. Claude 3.5 is your software developer. GPT o3 is your PhD intern. Donāt mix up the job descriptions.
Bengaluru woman books Uber, gets... Swiggy launches āDropsā for limit... US engineer quits ā¹1 cr job to bu... Download the medial app to read full posts, comments and news.
