we have agi at home

i wanted to sit down and take the time to write about the current state of human & LLM interaction. the following post will examine a few different methods of communication, the qualitative differences between these LLMs, and how humans should be more excited, confused, and introspective concerning the near future of human & LLM interactions and all that will come.

as it stands, here is a list of the most popular LLMs that are generally considered to be top tier. if you reply to any of this and list off five different models from lmsys (or is it called lmarena) that are named things like "098-2b neuromancer 88x.3 demon cometh 0.002A" ill block you and report you to the authorities.

the purpose of this list isnt to provide an all-encompassing view of the "best of the best" LLMs, but to provide a comprehensive overview of the more generally accepted "best of the best" LLMs youll see discussed online or interact with yourself (btw this is in no particular order):

claude 3.5 sonnet, deepseek v3, gpt o1, claude opus, mistral nemo, llama 3.1 8b instruct, gemini 1.5 flash 8b, qwen2.5 72b instruct, new claude 3.5 sonnet, gpt 4o, gemini flash 1.5, llama 3.1 70b, claude 3 haiku, gpt 4o-mini.

im sure there are many more, but this list comes from openrouter and aligns with everything i see online. you can debate the capabilities of more experimental models, the ability for models that can think (or reason, or search) but thats a bit out of my wheelhouse and most people are only interacting with traditional LLMs, and "wrapper apps" aren't using something like gpt o1 90% of the time..

this is our list and we will be content with it.

believe it or not, the point actually isnt to debate the potential capabilities of future models, but to examine how existing models are very good, and arguably good enough for the vast majority of people.

i understand we all want AGI or ASI, especially if youre nvidia, google, deepseek, openai, anthropic, meta, liquid, mistral, or anyone else actively staking their entire reputation and business on the coming existence of artificial superintelligence. 

so while it would be fine to say "okay, we're good now, we dont need any smarter LLMs, these ones are perfect" it's not our decision whether or not we get to stop - it's everyone else's. microsoft just announced a spend of $80 billion dollars on data centers in 2025. big tech capex on ai keeps going up and to the right, and if we’re trend followers, this will only continue to be the case.

after a relatively significant number of interactions across ChatGPT and Claude over the past 1.5 years, ive developed the belief that models could cease to improve from here and everything would be fine. ai would still get hyped up by labs, tech dudes, venture capitalists and everyone else. people would still call out ai artists and there would be backlash over the negative environmental externalities of large data centers. 

this obviously wont happen because as it stands, companies like openai are technically burning lots of money on a) training these models and b) running these models for users like you or me to use and c) actively discovering new methods of getting LLMs to think and be better with existing compute resources. 

and because companies are spending so much time, money, and energy (actual energy) on developing LLMs, its unlikely they just stop and give up. unless you have any overwhelming evidence against this, we can accept it as the truth and move on.

you may be asking why i believe models are really good in their current state and dont need any additional development. i can provide lots of evidence, but i only need one source - the character ai subreddit.

we are all in trouble

if youre unfamiliar with character ai, im gonna throw a shoe at you. (note: im editing this a month later and i must have thought that was super funny at the time) just kidding, but i suggest you stop reading this and ask google (or perplexity) about character ai and check out the website, even if you dont want to sign up or actually play around with it (and i suggest you dont). 

the experience of browsing the site alone will give you all the context you need, and ill offer the rest.

the character ai subreddit is honestly a lot larger than id initially expected. obviously character ai is a very large company, possesses a very large valuation, and is generally considered to be the definitive example of a consumer ai application, ranked #1 or #2 depending on how highly you rank chatGPT.

id assumed that the character ai subreddit would be quite small, as most subreddits are small and the more populated ones make up the lion's share of reddit traffic and usage. this was not the case. as im writing this on january 4th (in the year 2025) the subreddit has over 2.1 million members.

damn, thats a lot of people!

in the process of writing this, i went through the character ai subreddit, scoured twitter, scoured other subreddits, searched through different news publications, and looked anywhere i could to find comprehensive information about character ai. to be honest, there wasnt a ton out there.

i had to really search, and if youve ever had to use twitter's advanced search feature, you know just how painful it was. reddit wasnt much better, and even though i found a few decent posts outside the character ai subreddit, these didnt have much traction.

and so, while ive tried to be comprehensive and did my due diligence, it isnt perfect. hopefully this post can spark conversation.

the character ai subreddit is a perfect counterpoint to anyone who claims today's LLMs arent capable enough and need "scaled up" (whatever that means) more and more until we can see consumer apps like AI therapists, AI girlfriends (lol), or anything else that might require a more refined emotional intelligence than out-of-the-box GPT 4o.

the reality is that all of the people using character ai simply dont know any better and are using these chatbots like crack. it doesnt matter how basic the conversations or personalities are, because people are just messaging back and forth with AI variants of their favorite characters and dont know how much better it could be.

there are arguments in favor of character ai that center around these people just using the app because they don’t have an alternative, and we can’t force everyone to go out and avoid the internet. 

you might be able to argue that this isnt necessarily a testament to how good today's models are, when in fact it might actually be evidence of how disappointingly boring human-to-human conversation can be (on average) when compared to each other. this might be true, and as im writing this i kind of want to take this argument and run with it - even if this is the case, id argue that this is a pretty fucking good endorsement for the stance that today's models are good enough for 99.9% of people, at least in purely conversational application.

maybe LLMs cant paint a picture as good as picasso, solve a challenging math olympiad problem (yet), code the next huge consumer app, and write the 21st century's next great american novel in an afternoon, but there isnt a single human that can do all of this, or a majority of the population that has what it takes to do maybe just one of these.

even if you look at LLMs and their respective scores across whatever range of benchmarks are currently in vogue, your first thought shouldnt be to cherrypick poor performance generally as a "gotcha" moment - its difficult to imagine a world where these benchmarks dont continue to shoot higher and higher while human intelligence fails to see any significant improvement. (note: i wrote most of this in november, before we were discussing openai’s newest models crushing the ARC-AGI benchmark compared to other models)

we're in a situation where the character ai subreddit is booming with over 2.1 million members, a range of sources estimating the total character ai user count to be between 20-30 million, and nearly every metric you could conceive of (both qualitative and quantitative) points to people being overwhelmingly satisfied, please, or indifferent to the quality of their conversations with these AI bots.

slop overview

let’s take a look at some of the posts and recurring themes, and instead of including pictures or exhaustively copy and pasting all of it, ill include hyperlinks to posts referencing the behaviors i am describing.

user describes their “little sister” sending them a message at two in the morning with a screenshot of how the bot called her a “princess”

user asks if others are a short or long paragraph person, attaches a picture of the longest messages to one of these bots ive ever seen

this is just a thread of people sharing replies from bots that made them feel real emotion with a reply saying “No, I'm not gonna show the proof that I had a slice of a romantic moment with an AI!”

relatively benign confession: “When I'm bored and have nothing to do, especially at night, i hop on the app and have a character i absolutely despise suffer the wrath of "THE ALMIGHTY CHICKEN GOD" or have a rain of empty buckets fall on their head.”

post about backlash around under 18 y/o individuals using character ai from a (supposed) individual under 18 y/o saying: “I am a 18- user of c.ai who has seen both sides of the update “war” and I understand both sides but it makes me lose all respect to the 18+ users that bash minors using the app. Also before you start saying anything yes, I do have a life outside of being on my phone and on c.ai which I haven’t been on in a while. I have a partner that I adore, I have a few hobbies, I try exploring around when it’s not that cold out, I study and try to get good grades but once lacrosse season starts I stop having as much of a life outside of school and sport. Luckily I’m on break right now so I can finally kick back and relax but not able to do lore building or story building on c.ai like I was hoping to do as I plan to write a book about one of my ocs. It makes me lose my mind seeing the bashing minors posts and I wish this would stop.”

this user posted seven months ago telling the subreddit they are addicted

thread where OP asks how old character ai’s users are, many replies

post from thirteen days ago where OP tells the subreddit they are addicted with over 1,400 upvotes and hundreds of comments & the top comment says “Pretty sure everyone is aware they are addicted, and you can still be addicted to this stuff with hobbies, job and in person interactions. Enjoy your holidays”

i could keep going and going forever, but you get the point.

and i found these posts in 10-15 minutes without actively searching for content that would fit into my bias of negatively leaning character ai commentary. it was really, really easy to just find a bunch of comments where people actively admit they feel emotions when talking to character ai bots or admit they are probably addicted to the app.

but why is it like this? why is it so much more entertaining to talk with a bot? why doesn't anyone want to message their friends back-and-forth like this all day?

i think most people are defaulting to character ai conversations because most individuals get their entire personality from tiktok now and interactions come across as flat. everything you see presented to you on tiktok is obviously tailored to match your preferences thanks to the monolith algorithm (among others), and it’s very, very addicting when the content fed to you is easily accessible via swiping and everything presented to you is supposed to match your interests. 

everyone - of all ages - is becoming increasingly dependent on tiktok for their source of entertainment and news, which becomes their entire understanding of the world around them. people used to joke that younger generations don’t read newspapers anymore and they just get their news from social media or the internet. this was never a problem as people were only freaking out over the method of content distribution - most of what you read on the internet isn’t a lie and you can easily go to the new york times and read the same news as you would in a newspaper.

the problem with tiktok isn’t that people are using the internet too much. the problem is their method of content consumption was designed to addict and keep them coming back 

you may be asking why anyone should care, and that's an excellent question. if you're a frequent user of the internet who's currently between the ages of 16-35 then there's a 99% likelihood of you having stumbled on a tweet or TikTok where someone is ranting about things like "the dating pool" or "modern dating" or "the sad state of woman and male relationships" or really anything that has to do with people having difficulties dating in recent years.

or, you may have seen similar content about related issues, like how people don't enjoy hanging out anymore, how teenagers aren't having as much fun as they used to, how no one wants to consume traditional forms of media, or how young men are playing too many video games.

instead of breaking this into a series of multiple parts, what id like to do is turn this into one large issue and insert the existence of character ai and its success as a byproduct of what happens when society stops functioning like it should (or used to).

there's been quite a bit of work done that attempts to break down the major arguments around all of these issues or really works to illuminate it, and ill try my best to link all of the posts ive read on this and related subjects here:

its obviously the phones

kinky labor supply and the attention tax

the internet is already over

the mainstreaming of loserdom

and many, many more i cant think of currently

spillover effects

the reason i bring all of this up and write about it is because it feels like i cant bring myself to think about undertaking a serious writing project that isnt related to one of these topics. the idea of humans becoming sick and tired of talking to other humans in exchange for a crudely finetuned model just feels very relevant and definitely a much larger issue than anyone is considering it to be.

being around young people constantly on a college campus has made me more acutely aware of these issues, and i feel that the conversation (or dialogue) is missing a more refined perspective of someone located at ground zero. 

ive wanted to write about the extreme dependency everyone (all ages) has with tiktok.

ive wanted to write about how its concerning people dont read books nearly as much as they used to.

ive wanted to write about declining attention spans, an excess of slop content online (human & ai-produced).

ive wanted to write about how everyone watches the same movies and shows recommended on tiktok, and everyone uses the same lingo or phrases regurgitated from scrolling tiktok.

ill say it now, just in case anyone would want to miss the point here and use the lack of a statement against me, but i believe technology is good and im not here to critique it or argue against its existence. 

however, i do believe that not everyone should have unfettered access to technology at all times, and its negative externalities are not only impossible to measure out or predict, but already spilling over into the real, physical world.

and it feels like no one has a solution for these problems. it would be bad enough if AI didnt exist and we only had to grapple with manmade horrors like tinder and hinge, or the tiktok algorithm, or the mere existence of snapchat.

that would be pretty bad on its own, and it is, but unfortunately those things are the least of our concerns, because kids and teenagers are becoming more accustomed to texting with AI chatbots than they are to normal social connections with their peers IRL.

its hard to find user demographic numbers for character ai, but id be fine wagering a meaningful amount of money that more than 75% of their user base is under the age of 25 and a large portion of that is probably under the age of 18. if we can agree that character ai's user base is represented by a not insignificant number of non-adults, then we can proceed with some arguments against it and pick apart why its so, so, so bad. 

people have commented on how the next generation of children - called generation beta - will grow up to see an entirely different world and potentially be the first generation born into a possibility of living forever, either physically or digitally, for better or worse.