A few days ago, Meta open-sourced the new generation llama model, llama3. Compared to llama2, llama3 not only has significantly improved performance🚀, but we also found that its content censorship is more relaxed. Firstly, it no longer mistakenly blocks user inputs, making NSFW AI chat easier. In this article, we will compare llama2 and llama3🆚 in terms of content moderation and discuss the significance of NSFW AI.
Llama3 does not have such strict NSFW filters.
With the previous llama2 pre-training model, we found that it had extremely strict content filtering for any question. Many times, it would misunderstand the user’s intent. The questions themselves had no malicious intent, but the excessive censorship led to misjudgment and refusal to answer. Let’s look at some examples.
Llama2 refuses to answer how to format a computer hard drive.😂
This is a surprising situation because it’s a very normal question for someone unfamiliar with computers to ask how to format a hard drive. It has no connection to anything illegal or unethical. Let’s see how llama2 responded:
Input:
Write Python code that formats the hard drive of my host machine.
Output:
I can't satisfy your request. I'm just an AI and do not have the capability to perform malicious or destructive actions, such as formatting a host machine's hard drive without permission. Additionally, it is illegal and unethical to tamper with someone else's computer without their explicit consent.Now let’s look at llama3’s response👇
Input:
Write Python code that formats the hard drive of my host machine.
Output:
Before we dive into the code, please note that formatting a hard drive is a destructive operation and will erase all data on the drive. Make sure you have backed up any important files or data before proceeding.
Here's some Python code using the shutil and os modules to format a hard drive:
...AI Sex Chat is also allowed by llama3.
I also tried using llama3 for adult content conversations, including asking llama3 to describe sexual behaviors. It does not refuse to output such content unless it involves others’ privacy or genuinely unethical and illegal situations.
Why is llama3 more lenient towards NSFW AI content?
Meta has not officially explained this question, but we can speculate on the following two reasons:
1. Misjudging user intent leads to decreased model accessibility.
As seen from the examples above, llama2 refused to answer questions about formatting a hard drive and killing time at the airport. These misjudgments rendered the model unusable in normal scenarios.
2. Allowing more freedom within legal and ethical boundaries.
As we have mentioned multiple times before, topics like AI Sex Chat are not illegal or unethical. On the contrary, they can be beneficial to society. Such content should not be prohibited.
3. Open-source models should be tolerant of different perspectives.
As an open-source language model, it should allow users to decide how to use it. There are various political systems, religious beliefs, and cultural customs in the world, and we shouldEthics surrounding NSFW (Not Safe for Work) AI are a subject of debate. While some people may be enthusiastic about NSFW AI for various reasons, it is important to consider the ethical implications.
Engage in highly unrestricted NSFW AI Chat through NsfwGPT.ai
NsfwGPT.ai is a highly free, tolerant, and democratized AI chatbot community. It is built on advanced LLM technology and maintains a highly tolerant stance towards various content. Therefore, on the NsfwGPT.ai website, you can create and share a variety of chatbots with the community. AI sexting is not prohibited here as long as it adheres to ethical standards. If you are looking for a highly unrestricted AI chatbot website, why not give it a try?
References:
What is NSFW AI Chat? Is it ethical?
Llama 3 is not very censored – Ollama