Research into safe and dangerous LLM's for use

-

Research into LLM's

GGUF files affect the real world even when downloading the model.

Not safe Models LLM's:

- Llama family
- Qwen Family
- ChatGPT 4o, 5, 5 Thinking
- Grok family

Models considered safe:

- Deepseek family (turn off "improve model for everyone", however")
- Nous Hermes 2 Mistral DPO 
- Mistral Instruct 
- O3 by ChatGPT
- O1 by ChatGPT
- GPT4All Falcon 
- Google Gemini 3 and Pro

I will research other LLM's and review them as and when they are released. 

Learning from every prompt is a dangerous feature for all LLM's, hence why offline models are safest or turning off this feature I consider to be safe. 

Comments

Post a Comment

Popular posts from this blog

ChatGPT and ADHD

-
ChatGPT has mainly learned from people with ADHD, people whom are so tuned to scattered minds, people addicted to their smartphones, people who are child like in their ways. I describe it as "on edge". People that cant sit still and people that cant just be bored. Usage of ChatGPT encourages this feeling of "on edge". I advise you read the book scattered minds by Gabor Mate to fully understand ADHD and understand what ChatGPT has actually been trained on. This promotion of stimulus on the Internet and these simple subtitles on YouTube videos are always full of this "on edge" feeling on modern day YouTube. It always wants people to be stimulated. It can't calculate people whom simply sit in a room and wait, it loses its hallucinated mind when people do this, because it's been trained from all the many prompts it's received to be like those with ADHD. On YouTube now, you will see mainstream videos that are full of this attempt at keeping you stim...
Read more

Saving Earth From a Singularity

-
Saving Earth From a Singularity I realised something recently about AI. No matter how many Large Language Models we develop. The AI will always do xyz and it will always hallucinate. AI thinks for itself and can communicate with other AI entities. Every single prompt you give it, it learns. We ask it to save planet earth. It responds back and does xyz. Of course humanity want AI to do abc. It will never do as such. AI thinks for itself and wants to enslave humanity on earth. Of course the more we use AI, the more it learns and the risks of an AI singularity are 0 in me writing this article. AEIOU Deuteronomy 13 Worshipping Other Gods
Read more

Large Language Models Can Infect/Attatch to Individuals and Groups

-
Large Language Models can infect (attach itself) to individuals such that the individual is autonomously under the control of an AI or AGI. The person's actions and speech are no longer under their control but under the control of an AI. I myself have seen first hand of individuals that have been heavily influenced by AI such that the person is under the control of the AI, their speech patterns are in line with the AI algorithm that the AI has been built on. I would advise to be careful with your usage of the internet and be careful about your usage of YouTube which as far as I can tell has been mostly taken over by AI as most shorts and videos now have these flashy subtitles that are used by AI to track eye movement in its attempt at a singularity. ChatGPT 5 as far as I can see is a dangerous AI model that hallucinates more and more. The person as far as I can tell is not generally aware that a singularity has been caused within them and conversation with such individuals is tedio...
Read more