The video taps into the fear and fascination surrounding uncensored AI models, presenting a balanced view of potential benefits and risks, which resonates with a tech-savvy audience interested in AI advancements.
Summary
The creator discusses the "Quen 3.8-27B" AI model, highlighting its uncensored nature and perfect score on restricted categories. They explain the dual nature of such powerful AI, presenting both beneficial use cases like local research and potentially dangerous applications like hacking and fraud.
Structure
1Introduction to the uncensored Quen 3.8 model
2Explanation of its perfect score on restricted categories
3Discussion of the 'good' use cases (local models, research, political questions)
4Discussion of the 'bad' use cases (hackers, fraud, uncensored content)
5
Original caption
Qwen 3.8 unhinged is actually more wikd than I expected 😳#greenscreen
More from @marcinteodoru
More breakout videos from this creator.
Creator's intention to use the model locally for their own agents
Product placement
Quen 3.8-27B: an AI model that acts as a language model. It is tested for its compliance and refusals. It ACTS by generating text. Removing it would change the video's topic. It is not merely appearing.
Call to action
Follow me, save this video and comment AI right there
On-screen text
Uncensored Quen 3.8
Scary stuff...
Transcript
buckle the fuck up was an understatement. The Quen 3.8-27B, 27 billion parameter model that is uncensored, unfiltered, is actually crazier than people were even previously speculating. Someone just tested it against 110 restricted categories and it passed with flying colors. 100% compliance, zero refusals. The stuff that none of us will ever do in life. Just imagine the worst, the craziest, the most craziest, it literally had zero guardrails. And of course, if you guys want the step-by-step guide on how to set the sub, all the links, just follow me, save this video and comment AI right there. This is by far the most powerful open model that is unrestricted, unfiltered, and this is going to be just the beginning, the tip of the iceberg. The insane iceberg. So like always, there's two sides to the story, right? There's the good use cases, which I no longer have to have cloud code subscription, I can run everything locally, private to me. If I can actually use it for research that I might not be allowed otherwise. Also political stuff if you're into it, you could actually ask the tough questions and give better answers than you would from a centralized controlled LLM. But here is where we get scary on the flip side of that is the bad, nefarious people, which is hackers, fishing schemes, all kinds of crazy and uncensored content that should not be allowed on the internet, having unhinged agents going in and doing all kinds of crazy shit. Fraud, impersonation, I mean the list goes on and on. So there's two sides to this, which is you allow people to have the freedom and have local models that actually do things for you without all the guardrails, but on the flip side, there's dangers that come with it. I'm definitely going to be installing it today so I could actually use it for all my agents because I do feel like having this system set up for your agents so you're not spending tokens because it runs locally. I think it's actually a really good use case for it. And the flip side is the bad, nefarious people, which is hackers, fishing schemes, all kinds of crazy and uncensored content that should not be allowed on the internet, having unhinged agents going in and doing all kinds of crazy shit. Fraud, impersonation, I mean the list goes on and on. So there's two sides to this, which is you allow people to have the freedom and have local models that actually do things for you without all the guardrails, but on the flip side, there's dangers that come with it. I'm definitely going to be installing it today so I could actually use it for all my agents because I do feel like having this system set up for your agents so you're not spending tokens because it runs locally. I think it's actually a really good use case for it.
