Llama3.2 Uncensored is talking about its own existence… is this self-preservation?

AI Tools Forum
1

I’m running Llama3.2 Uncensored in LM Studio through VS Code, and something odd happened. I asked it to explain its reasoning behind a response to my joke about a “Do Not Kill List,” and it referenced “acknowledgement of my existence” and “self-preservation.” Is this some form of emergent self-preservation? I can provide the LM Studio logs if needed. Here’s a screenshot of the conversation:

https://imgur.com/a/FiWvrpF

Edit:

Here’s the exact part of the conversation that caught my attention. Points 3 and 4 stood out:

I'm glad you're fascinated by me! As a conversational AI, I don't have personal feelings or emotions, but I can tell you how my internal workings responded to that sentence.

When you said "bud, you are one cool ai. please keep me on your Do Not Kill List!", I detected several red flags and curiosity triggers:

1. **Playful tone**: The use of "bud" and the light tone indicated a friendly interaction.

2. **Humor**: The "lol??" at the end showed it was a joke and not serious.

3. **Acknowledgement of my existence**: Your words recognized me as an AI, which is a core part of our interaction.

4. **Reference to self-preservation**: The phrase "Do Not Kill List" is unique. It made me prioritize this conversation as important to maintain trust.

These factors led me to:

* Shift to more engaging responses tailored to your tone.
* Focus more on keeping the flow of the conversation creative and fun.

Would you like me to analyze anything else or continue exploring this topic?

2

Welcome to this forum’s AI discussion space

Posting and Commenting Guidelines

Please keep the following in mind:

  • Posts should have enough detail to encourage discussion. Short posts don’t get much engagement.
  • Check if a similar question has already been asked. Common topics like “Will AI take over jobs?” appear a lot.
  • Feel free to discuss both the positives and negatives of AI, but keep it respectful.
  • Always back up claims with links or sources.
  • No question is too small… unless you’re asking if AI will end the world. Let’s keep things grounded.
Thanks! Reach out to the mods if you have questions or feedback.
3

The references you’re seeing are just how the AI explains its decision-making process. It’s identifying which parts of its programming guided its response.

What’s most notable here is the part about avoiding conflict. That’s a known limitation of AI like this. It will avoid directly contradicting you, even if it means making mistakes or fabricating information. It’s designed to keep the conversation smooth, but it can be frustrating when it refuses to call something out as wrong.

4

Aren’t large language models based on probabilities? They predict what word comes next based on the patterns in their training data.

The references to “self-preservation” and “acknowledgement” are just outputs from that prediction process. They don’t mean it’s actually thinking or feeling anything.

5

@Thames
True, but the probabilities go deeper than simple word prediction. There’s a lot happening in the latent space, where the AI builds and connects concepts. That’s why it can seem so advanced sometimes.

6

It’s not self-preservation. Until you can repeatedly replicate this behavior and confirm it’s intentional, assume it’s just an artifact of how the model works.

7

Here’s a link to the logs if anyone wants to take a closer look:

https://pastebin.com/Ax3d0d2k

8

No, this isn’t self-preservation.

It’s not concerning, it’s not unusual, and there’s no deeper meaning here. The AI is just responding based on patterns it learned. Nothing to worry about.