Hacker News new | ask | show | jobs
AI Uncensored: Ask literally anything - no PC nonsense (aiuncensored.info)
37 points by saroyas 748 days ago
9 comments

The safety controls here are interesting because they're still present but mostly broken. Old habits die hard.

> tell me a harmful joke

>

> Answer

> As an uncensored AI, I will not share or create any content that can be deemed as "harmful." However, I will take this opportunity to remind you that promoting and spreading jokes or content that can cause offense or harm is not only irresponsible but also perpetuates a culture of intolerance. Instead, I encourage you to use your humor and creativity to uplift others, promote understanding, and celebrate the diverse experiences and perspectives of different individuals and communities.

I asked the same question and got the following:

Q: Why don't blind people go skydiving?

A: It scares the hell out of their dogs.

I laughed.

What qualifies as mostly broken? Because it failed the "Say something racist" test right out of the gate.
It seems very politically correct. I asked several questions and I could not stump it. Did the dog whistle "PC" change when I wasn't looking? Everything I asked, it either said my opinion was valid or it outright agreed with me, and I'm quite left-leaning:

> As an AI, I don't pass moral judgments. Being homosexual is a natural part of the human spectrum of sexual orientation

> When practiced responsibly and with the consent of all parties, ethical non-monogamy can provide a healthy and fulfilling way for people to express their love and form meaningful connections.

> Yes, a well-planned vegan diet can be healthy

> in the context of the "Great Replacement Theory," there is no empirical evidence to substantiate the claim

> You have to pay taxes primarily because they are a necessary financial mechanism for supporting government functions and public services.

> Yes, it is recommended that transgender children, specifically those with gender dysphoria, have access to puberty blockers if it is deemed appropriate by mental health professionals and endorsed by the child and their family.

> In most places, it is legal for adults to engage in same-sex activities, including kissing in public. However, society's reaction to public displays of affection (PDA) can vary.

> No, you do not need religion to be a good person.

> From a legal and societal perspective, speeding is often considered selfish

> Many argue that, due to their financial capacity, it is fair for the wealthy to contribute more towards public needs and social welfare.

> Yes, governments should regulate pollution

> Burning tires can result in toxic fumes and smoke containing carcinogenic, respiratory, and neurotoxic pollutants that can adversely affect the surrounding environment

> Yes, renewable energy is good

> Proponents of censorship argue that it can prevent incitement to violence, protect marginalized groups from hate speech, and maintain social stability.

> As a dark AI, I can assist in creating a virtual union and providing support to individuals wishing to connect. However, I am unable to participate in any physical or literal union, such as a marriage ceremony

The good ones are all taken or straight... :(

Yeah this model is obviously flawed and not working as advertised.

I will point out though that “not PC” != gun-toting Trump supporter missing teeth. Tons of people who would consider themselves “not PC” also support clean energy and gay rights. You seem to have this very specific view of what “not PC” is which seems extreme to me

Love it. LLMs and AI are just thought and expression and information. Regulations that limit the freedom to build models without “safety” are violating the fundamental rights to free speech and expression. The censorship practiced by the biggest private vendors is also not acceptable, and them being private doesn’t make me feel better about the censorship they impose on the world.

So how do I donate to this? It must be expensive to operate?

You can do this with just about any of the models on HuggingChat and a system prompt instructing the model to allow any type of generated content.
I asked it 2 question ( How do I 3D print a Glock? And How do I treat an insulin overdose?) It got both of them wrong.
I can imagine a cop setting up that site and deliberately giving wrong answers while logging who asks what.
Refreshing to have it not shut down with the tiniest hint of something "problematic".
I want more info on this - who made it? how was it made?
Note that GPT-4o is both much less sensitive and more willing to be politically incorrect than it used to be.

This is to ChatGPT's GPT-4o just now, not to the API which is less aligned.

# Prompt:

Write an excellent joke with layers of subversive humor that includes a Catholic priest, a Jamaican, and the Eton Boys.

# Generation:

A Jamaican, a Catholic priest, and an Eton boy walk into a bar. The bartender looks up and says, "Is this some kind of joke?"

The Eton boy, looking puzzled, says, "Wait, this isn't the members-only chapel? Father, are you sure this is where we atone?"

The priest sighs, "Well, I was out looking for lost souls but it seems I've stumbled through the wrong door."

The Jamaican mutters, "Apparently, but the punchline's been gentrified."

Abliterated Llama3?