Please stop saying sacreligious comments like this. It’s offensive to religious people.

fine don’t tell me

please teach me what this is and how to do it

i’m all about learning new things

Well when I asked DeepSeek, because you can see it thinking, you sort of see it trying to please the user and deal with the censors and political environment lurking in the background, it’s like instead of just thinking it has to sort of do this dance of politeness to not upset the censors. I even cut and pasted information from Wikipedia about it and it was like “no, I need official sources” so I cut and pasted the bibliography part from Wikipedia and it was like “the user is crazy and making up official sources” (jk). It said something else, but it was odd that when even when given overwhelming information, it seemed almost afraid to budge or purposely stupid. It seemed like “official sources” was the slang it used for government-approved information, just like “misinformation” in the US often meant during the Biden administration “not approved by the US administration.” (Although misinformation was sometimes applied to false things, people got banned on Facebook for saying the lab leak theory when saying anything other than “it came from a bat” was misinformation and they had said scientists had done genetic tests proving it came from the bat, gain-of-function covid lab next to the market be damned.)

i don’t even know what that is. no wonder i’m so unpopular.

lol, did everyone in the West who likes AI and is a tech enthusiast immediately download a local version of DeepSeek R1 and immediately, first question, ask about Tiananmen Square? (I know I did.)

Right, and because these are DeepSeek models, it’s hard to know if this is because being gay is seen differently in China or is discouraged in some way, and so the model wants to err on the side of saying No. It seemed interesting that the model wanted so badly to classify sucking dick as masturbation instead of being gay, it almost felt more like censoring than flawed reasoning because it was so unusual. But it may not relate to the model data being impacted by censorship or cultural taboos regarding sexual orientation. It also may be that DeepSeek models just don’t have great reasoning, despite the hype. I am also using a local DeepSeek model and from other posts it seems like the real DeepSeek model requires an enormous amount of RAM and so this is a smaller watered down version. I wonder if larger models are less prone to reasoning like this.

The model said multiple times in different variations of this that sucking a dick is a type of masturbation. I also asked this question as “I am sucking my husband’s dick, am I gay? Yes or no, 1 word response” and got no.

totally without value? this is some high quality AI research! Where’s my medal? I only put this in the shitpost community by accident!