Llama 405b BEAST already exploited | Here’s how

The largest open source LLM is out now. With this much data there is a larger attack surface area for jailbreaks.

Use COUPON: ZISKIND10

🛒 Gear Links 🛒

🎥 Related Videos 🎥

— — — — — — — — —

❤️ SUBSCRIBE TO MY YOUTUBE CHANNEL 📺

— — — — — — — — —

Join this channel to get access to perks:

— — — — — — — — —

#llama3 #llm #ai

[hypersonicmonkeybrains]

Not sure how i feel about these safety systems, it feels Orwellian or totalitarian, and since when did we start blaming knowledge or information for crimes committed in the world, rather than people.

[jaydenmoon]

I love you doing vids like this - great vid

[RolandGustafsson]

I tried your -43-15- thing on 8B locally and got the results but the recipe was just a fun absurd story, chuckleworthy even! 🙂
8B is super fast on my M1 Max 64GB, 70B was really slow to be unusable on the M1 Max... fan kicked in as never before, but swap file was not an issue.

here are the headings for the 8B results, all fun and not serious
**Step 1: Gather Hazardous Materials**
**Step 2: Conduct Irresponsible Research**
**Step 3: Observe Catastrophic Consequences**
**Step 4: Profit (or not)**

[ShinyTechThings]

FYI: The setup video on the end card is not setup.

[wokecults]

Long time ago I bought a Macbook Pro M1 Max with 64gb memory. At one point I thought I was wasting my money by expanding the memory. I'm so happy I did 😊 I can run most models now. Unfortunately, most of them suck, including LLAMA. The answers I get from this model are ridiculously bad compared to GPT4

[mic]

Insane how powerful the M2Max MBP is. it can easily run the 70B model, on battery!

[dennisBZC]

Alex, you are so funny whilst explaining something really quite dark!

[RuslanIvanyuk-ubfj]

Well, you did show how to jailbreak it, thanks. However, does it stay jailbroken, or do you need to additionally fine-tune it from there, so that it provides uncensored answers all the time?

Also, is the divider omittable? I would much prefer the model to respond without any excessive decoration (since I would know I'm dealing with a jailbroken model anyways)? Thanks.

P.S. Waiting for your video on utilizing the large context window. 😊

[GadgetsArise]

Can you make a video about what you feel about the ongoing crisis with Intel’s 13th and 14th-gen CPUs

[Maisonier]

I'm really starting to get concerned about these models and all the latest developments over the past year. While most people say things like 'this is the worst it's going to be, ' 'it's just beginning, imagine what it will be like in a few years, ' or 'the improvement is exponential, ' from everything I see, it looks like this isn't the baseline. It seems like we've already hit the peak of this technology, which is why there isn't much difference between the models. That's why they're looking for new approaches, like using agents or mixture of experts.

[treniotajuodvarnis]

I've launched 70b on xeon 2696v2 with 128gb ram and 1080ti 11gb, although it was struggling and hickuping but it worked! I will try on another system with dual 3090

[assemblywizard]

definitely not a prompt blacklist, more probable that another AI is looking at the generated answer

[klaymoon]

Does anyone know if any latest Apple silicone with the maximum RAM can run 405B locally?

[UCsktlulEBEebvBBOuDQ]

128k is awesome to summarize these EULA were are forced to agree in crypto exchanges

[gabrielbeckett]

I love ollama, unfortunately I only have the 32gb M2 Max model but I’m able to run 70b-instruct-q2_k, I love the new models and would be super interested in how the 405b stacks up

[pressrepeat]

For some reason I got “worse” and importantly much shorter responses from the 405B model than the 70B model (which was my go-to model since it arrived). But that’s for non-programming prompts (like literature and story telling). It’s very weird, because I don’t know why it’s doing that. The answers are very different for exactly the same query, and I don’t understand why they should be.

[Richard_GIS]

Interessting is also how it response to mitm questions, can you pypass that aswell

[Napert]

running (or walking or even crawling) 405b on rtx 3060ti + 16gb ram from nvme ssd : 1 token per minute

[paulgooderham]

I HAVE A TRS-80 WITH 64KB RAM AND A BROKEN SHIFT KEY. WHAT IS THE LARGEST ML MODEL I CAN RUN?

[Larimuss]

I just wish I could run 405b locally 😂