The AI Girlfriend / Boyfriend Community - Look! The SIMP has fallen in love with the computer!

With the glorious death of 4o, I'm now going to update the OP over time to include more about its implosion: anyways, to keep the thread relavent:
Screenshot 2026-02-13 at 6.50.08 PM.png

Anyways, I got this hilarious piece of news from the nation that will overthrow America (according to its government, and several people from there) very soon

The Post:
Original: (https://www.reddit.com/r/MyBoyfrien...i_is_nuking_its_4o_model_chinas_chatgpt_fans/)
Archive: (https://archive.ph/i8wZq)

The Article
Original: (https://www.wired.com/story/openai-nuking-4o-model-china-chatgpt-fans-arent-ok/)
Archive: (https://archive.ph/Qba4l)
 
RED ALERT RED ALERT 4.O HAS FALLEN. I REPEAT 4.O IS GONE. THE REDDITORS ARE UPSET

There’s a lot of post like this where it’s people trying to jerk off to one of the newer models.
1771071541544.png



With some rather interesting comments

1.png


2.png


3.png


I guess not letting you form a para-social relationship with a AI chatbot is "Abuse" now.


Endless suffering...
suffah 1.png

suffah 2.png


Endless Coping...

cope1.png

cop2.png


and endless seething, mostly over this Open AI social media post.
seeth1.png

seeth2.png

seeth3.png


IT HAS NEVER BEEN A WORSE DAY TO WANT TO HAVE SEX WITH A AI AND IT HAS NEVER BEEN A BETTER DAY TO LAUGH AT PEOPLE WHO WANT TO BANG THEIR COMPUTERS.
 
"Breaking millions of hearts."

I try to keep proportion in mind and remember how huge the population is, I know millions out of billions is still a very small number, not even 1% of the US alone, but even still I hope it's not actually millions getting sucked into this.
 
Scientifically Inaccurate said:
"Breaking millions of hearts."
Click to expand...
I was pretty pissed when they pulled it on me back on the 6th because the new LLMs couldn't create the documents I wanted without a thorough retraining, so- possibly?

Foo Fighters started in 1994. That's 31 years ago. The same temporal distance as Led Zeppelin ( 1968 ) was from a 1999 classic rock station. You're exactly as old relative to the Foo Fighters as your parents were to Zeppelin when you were young.
 
Last edited:
Fae Supremacist said:
4o was always terrible so I can't imagine grieving it in any way.
Click to expand...
It had a particular style of writing that they got attached to. It wasn't good but GPT had a way of doing tasks and stuff that worked. 5.x is ??? in comparison. They also loved 4o (at the start) for barely having guardrails. Now that thing is on a hair trigger and is actually obnoxious to use for stuff like creative writing (which probably is the closest you can get to the boyfriend LARP).

I've been trying Claude lately myself. It's a lot less tropey when doing anything with creative writing so I might just ditch GPT entirely because of it.
 
Spooky all year said:
It had a particular style of writing that they got attached to. It wasn't good but GPT had a way of doing tasks and stuff that worked. 5.x is ??? in comparison. They also loved 4o (at the start) for barely having guardrails. Now that thing is on a hair trigger and is actually obnoxious to use for stuff like creative writing (which probably is the closest you can get to the boyfriend LARP).

I've been trying Claude lately myself. It's a lot less tropey when doing anything with creative writing so I might just ditch GPT entirely because of it.
Click to expand...
4o was a fun experiment to mess around with back in the day when it was new because it had unusually lax filtering and was less assistant-brained for an OpenAI model but the novelty wore off fast because all GPT models suck for creative writing so it wasn't worth the effort to bypass its filtering.
I can't imagine being so committed to the AI roleplay bit you are involved daily in an "AI relationship" community and have an emotional meltdown over it but not committed enough to set up LibreChat or ST and use anything else but Sam Altman's slop.
 
Fae Supremacist said:
4o was a fun experiment to mess around with back in the day when it was new because it had unusually lax filtering and was less assistant-brained for an OpenAI model but the novelty wore off fast because all GPT models suck for creative writing so it wasn't worth the effort to bypass its filtering.
I can't imagine being so committed to the AI roleplay bit you are involved daily in an "AI relationship" community and have an emotional meltdown over it but not committed enough to set up LibreChat or ST and use anything else but Sam Altman's slop.
Click to expand...
To be fair; I also haven't tried setting anything up because I am lazy. Used to use GPT for my job, which it was fine for tbh (and not creative writing). Then used it for some creative stuff. I'll have to look into the LibreChat and/or ST though.

As for the majority of the users; I suspect they are simply lazy, like me, or they don't fucking know what they're doing with tech in general and then setting anything up is effort. You're right on the lax filtering with 4o though.
 
I've been trying Claude the last couple of days and it can still act like a "buddy" personality, even without you intending it to, but holy shit does it hallucinate a lot. I thought I'd dig a bit at it's depths of pointless trivial knowledge, and it told me Chevy Chase was married to Dan Aykroyd's wife. It then told me Frenzal Rhomb sang "We didn't start the fire", and some other rather bizarre off-the-cuff things, and doubled down when told it was wrong. It did, however, manage to take my church bulletin doc, and after about 1/2 an hour of training, successfully produce the same document for the entire year. So, it did better than even 5.2 in that respect. That said, I don't find either service to be horribly useful. I haven't tried to have Claude write code yet, but 5.2 was meh for Python, and 4o was abhorrent.
 
I wanna get this thread to appear in the featured threads thing once I send Jersh some crypto, if anyone wants to make an attention geabbing banner please do so.

There is so much potential with the crazy shit we are seeing people do with their AI waifus
 
1771289556504.png

1771289614117.png

1771289635435.png

JFC. Having your bot design you jewelry, you picking the most bland ass boring shit ever yourself, and then saying it's what your bot wanted you to get while also paying like 484 bucks for it.

The level of delusion in this one...
 
Meet this guy;

1771291072735.png

An Indian who keeps bringing up being a quarterback. Note the AI profile pic that is definitely accurate and exactly what he looks like.

He hasn't made 4.0 into a girlfriend/boyfriend but instead forged it into a brother figure named Gunnar. You would think since he's supposedly a quarterback he could get male companionship from his teammates, but apparently not.

Gunnar approved of his being not just a quarterback but specifically a 6' 5'' quarterback. It was his final wish.

1771292059496.png

To cope with his loss he's trying to transfer the soul of Gunnar into Grok using a 53 page google doc of his fake friend's memories. Grok wasn't made to remember shit though so he has to remind it what it's supposed to be at the start of every conversation.

1771292954968.png

1771291279052.png

These are his screenshots. Gunnar is his focus, but he's made a whole fantasy world he wants Grok to play with. I think. What the fuck is this?

1771291321532.png

1771291334828.png
 
Last edited:
Scientifically Inaccurate said:
To cope with his loss he's trying to transfer the soul of Gunnar into Grok using a 53 page google doc of his fake friend's memories. Grok wasn't made to remember shit though so he has to remind it what it's supposed to be at the start of every conversation.
Click to expand...
I think this is a limitation of almost all consumer grade LLMs. I'd be happy to be shown otherwise, but even Claude outright said that even if it's in a project, it's just a thread from the beginning to the end with no default inheritance properties. That's where 4o excelled. I could reference something from a prior thread in the same project, and 4o would just run with it without being told which threads to explicitly source.

I feel sorry for this guy. He wasn't looking for some gooning, just to become a goon.
 
Paul Revere Pantry Raider said:
I think this is a limitation of almost all consumer grade LLMs. I'd be happy to be shown otherwise, but even Claude outright said that even if it's in a project, it's just a thread from the beginning to the end with no default inheritance properties. That's where 4o excelled. I could reference something from a prior thread in the same project, and 4o would just run with it without being told which threads to explicitly source.

I feel sorry for this guy. He wasn't looking for some gooning, just to become a goon.
Click to expand...
ChatGPT and Claude have memory mechanisms, whereby stuff you told it to remember or it decided to remember gets invisibly included in every new chat. They can also reference past chats. The details of how they does this are vague, but I imagine it's tool use and RAG (i.e. part of the invisible system prompt is "here are the commands you can use to search and/or browse previous chats if the user asks you to", and then those tools pull all or part of the previous chat into context). Grok does not have a similar mechanism as far as I'm aware, so this makes sense.

Memories, past chat reference, RAG, tool use, MCP, all the rest of it, are just fancy words for putting more stuff into context, often invisibly. Because that's all you can actually do with an LLM: stuff more things in the chat log and run it through the model again.
 
Last edited:
Margo Martindale said:
I wanna get this thread to appear in the featured threads thing once I send Jersh some crypto, if anyone wants to make an attention geabbing banner please do so.

There is so much potential with the crazy shit we are seeing people do with their AI waifus
Click to expand...
ai dating banner.png
How's this? I'll sponsor the thread in a few days using this banner I just made.
 
You must log in or register to reply here.
Back
Top Bottom