This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
I have seen the AGI, and it is Gemini 2.5 Pro
If you go and look at Metaculus you’ll see that regardless of the recent breakthroughs like VEO3 and OpenAI’s “Ghiblification” (probably the first AI image system where accusing the outputs of being “slop” makes the accuser look unreasonable rather than denigrates the picture itself) all the “when AGI?” benchmarks have been uncharacteristically stubborn. The question asking about “weak” AGI has gone nowhere for two
weeksmonths while the median prediction on the question about full AGI has receded three years from 2031 to 2034.It looks like Scott’s AGI 2027 push has failed to convince the markets. For the informed person, AGI is coming “soon” but isn’t imminent. However I think that actually AGI is already here, is freely available to anyone with an internet connection and is called Gemini 2.5 Pro.
For those of us not in the know, at the moment you can access Gemini 2.5 Pro for free with no limits on Google’s AI studio right here: https://aistudio.google.com/prompts/new_chat ; yep, you heard that right, the literal best text model in the world according to the lmarena.ai leaderboard is available for free with no limits and plenty of customisation options too. They’re planning on connecting AI studio access to an API key soon so go and try it out for free right now while you can. No need to overpay for ChatGPT pro when you can use AI studio, and it’s a lot lot better than the Gemini you get via the dedicated app/webpage.
Our story begins a few days ago when I was expecting delivery of a bunch of antique chinese hand scroll paintings I had purchased. Following standard Chinese tradition where collectors would add their own personal seal in red ink to the work and seeing as these scrolls already had a bunch of other seal glyphs I wanted to add my own mark too. The only issue was that I didn’t have one.
This led to a rabbit hole where I spent a good portion of my Saturday learning about the different types of Chinese writing all the way from oracle bone script to modern simplified script and the different types of stones from which seal were made. Eventually after hours of research I decided I wanted a seal made from Shoushan stone written in Zhuànshū script. That was the easy part.
The real difficulty came in translating my name into Chinese. I, with a distinctly non Chinese name, don’t have an easy way to translate the sounds of my name into Chinese characters, which is made all the harder by the fact that pretty much all Chinese syllables end in a vowel (learning this involved even more background reading) even though my name has non-vowel ending syllables. Furthermore, as a mere mortal and not a Son of Heaven with a grand imperial seal, decorum dictated that my personal mark be only 4 characters and around 2cm*2cm, enough to be present but not prominent on the scroll.
All this led to further constraints on the characters to be put on my seal, they couldn’t be so complex that carving them on a small seal would be impossible, and yet I needed to get my name and surname as accurately onto it as possible. Naturally this involved a lot of trial and error and I think I tried over 100 different combinations before coming up with something that sort of (but not completely) worked.
There was one syllable for which I could not find any good Chinese match and after trying and rejecting about a dozen different choices I threw my hands up and decided to consult Gemini. It thought for about 15 seconds and immediately gave me an answer that was superior to literally everything I had tried before phonetically, however unfortunately was too complex for a small seal (it wouldn’t render on the website I was buying the seal from).
I told Gemini about my problem and hey ho, 15 seconds later another character, this time graphically much simpler but sounding (to my non-Chinese ears) exactly the same was present and this actually rendered properly. The trial and error system I was using didn’t even have this particular character as an option so no wonder I hadn’t found it. It also of its own volition asked me whether I wanted to give it my full name so it could give me characters for that. I obliged and, yes, its output mostly matched what I had but was even better for one of the other syllables.
I was honestly very impressed. This was no mean feat because it wasn’t just translating my name into Chinese characters but rather translating it into precisely 4 characters that are typographically simple enough to carve onto a small seal, and with just a few seconds of thought it had managed to do something that had taken me many hours of research with external aids and its answer was better than what I had come up with myself.
All this had involved quite a bit of back and forth with the model so out of curiosity at seeing how good it was at complex multi step tasks given in a single instruction I opened up a fresh chat and gave it 2-3 lines explaining my situation (need seal for marking artworks in my collection). Now I’m an AI believer so I thought it would be good enough to solve the problem, which it absolutely did (as well as giving me lots of good unprompted advice on the type of script and stone to use, which matched my earlier research) but it also pointed out that by tradition only the artist themselves mark the work with their full name, while collectors usually include the letter 藏 meaning “collection”.
It told me that it would be a Faux Pas to mark the artworks with just my name as that might imply I was the creator. Instead it gave me a 4 letter seal ending in 藏 where the first three letters sounded like my name. This was something that I hadn’t clocked at all in my hours of background reading and the absolute last thing I would ever want is to look like an uncultured swine poseur when showing the scrolls to someone who could actually read Chinese.
In the end the simple high level instruction to the AI gave me better final results than either me on my own or even me trying to guide the AI… It also prevented a potential big faux pas that I could have gone my whole life without realizing.
It reminded me of the old maxim that when you’re stuck on a task and contacting a SysAdmin you should tell them what your overall goal is rather than asking for a solution to the exact thing you’re stuck on because often there’s a better way to solve your big problem you’ve overlooked. In much the same way, the AI of 2025 has become good enough that you should just tell it your problem rather than ask for help when you get stuck.
Now yes, impressive performance on a single task doesn’t make AGI, that requires a bit more. However its excellent performance on the multilingual constrained translation task and general versatility across the tasks I’ve been using it for for the last few weeks (It’s now my AI of choice) means I see it as a full peer to the computer in Star Trek etc. It’s also completely multimodal these days, meaning I can (and have) just input random PDFs etc. or give it links to Youtube videos and it’ll process them no different to how a human would (but much faster). Funny how of all the futuristic tech in the Star Trek world, this is what humanity actually develops first…
Just last week I’d been talking to a guy who was preparing to sit the Oxford All Souls fellowship exam. These are a highly gruelling set of exams that All Souls College Oxford uses to elect two fellows each year out of a field of around 150. The candidates are normally humanities students who are nearing the end of their PhD/recently graduated. You can see examples of the questions e.g. the History students get asked here.
However the most unique and storied part of the fellowship exam (now sadly gone) was the single word essay. For this, candidates were given a card with a single word on it and then they had three hours to write “not more than six sides of paper” in response to that prompt. What better way to try out Gemini than give it a single word and see how well it is able to respond to it? Besides, back in 2023 Nathan Robinson (or Current Affairs fame) tried doing something very similar with ChatGPT on the questions from the general paper and it gave basically the worst answers in the world so we have something to compare with and marvel at how much tech has advanced in two short years.
In a reply to this post I’m pasting the exact prompt I used and the exact, unedited answer Gemini gave. Other than cranking up the temperature to 2 no other changes from the default settings were made. This is a one-shot answer so it’s not like I’m getting it to write multiple answers and selecting the best one, it’s literally the first output. I don’t know whether the answer is good enough to get Gemini 2.5 Pro elected All Souls Fellow, but it most certainly is a damn sight better than the essay I would have written, which is not something that could be said about the 2023 ChatGPT answers in the link above. It also passes for human written across all the major “AI detectors”. You should see the words and judge for yourself. Perhaps even compare this post, written by me, with the output of the AI and honestly ask yourself which you prefer?
Overall Gemini 2.5 Pro is an amazing writer and able to handle input and output no different to how a human would. The only thing missing is a corporeal presence but other than that if you showed what we have out there today to someone in the year 2005 they would absolutely agree that it is an Artificial General Intelligence under any reasonable definition of AGI. It’s only because of all the goalpost moving over the last few years that people have slowly become desensitized to chatbots that pass the Turing test.
So what can’t these systems do today? Well, for one they can’t faithfully imitate the BurdensomeCount™ style. I fed Gemini 2.5 Pro a copy of every single comment I’ve ever made here and gave it the title of this post, then asked it to generate the rest of the text. I think I did this over 10 times and not a single one of those times did the result pass the rigorous QC process I apply to all writing published under the BurdensomeCount™ name (the highest standards are maintained and only the best output is deemed worthy for your eyes, dear reader). Once or twice there were some interesting rhetorical flourishes I might integrate into future posts but no paragraph (or larger) sized structures fit to print as is. I guess I am safe from the AI yet.
In a way all this reminds me of the difference between competition coding and real life coding. At the moment the top systems are all able to hit benchmarks like “30th best coder in the world” etc. without too much difficulty but they are still nowhere near autonomous for the sorts of tasks a typical programmer works with on a daily basis managing large codebases etc.. Sure, when it comes to bite sized chunks of writing the AI is hard to beat, but when you start talking about a voice and a style built up over years of experience and refinement, well, that is lacking…
In the end, this last limitation might be the most humanizing thing about it. While Gemini 2.5 Pro can operate as an expert Sinologist, a cultural advisor, and a budding humanities scholar, it cannot yet capture a soul. It can generate text, but not a persona forged from a lifetime of experience. But to hold this against its claim to AGI is to miss the forest for one unique tree. Its failure to be me does not detract from its staggering ability to be almost everything else I need it to be. The 'general' in AGI was never about encompassing every niche human talent, but about a broad, powerful capability to reason, learn, and solve novel problems across domains—a test it passed when it saved me from a cultural faux pas I didn't even know I was about to make. My style, for now, remains my own, but this feels less like a bastion of human exceptionalism and more like a quaint footnote in the story of the powerful, alien mind that is already here, waiting for the rest of the world to catch up.
And:
Em-dash spotted. Thought you could pull a fast one on me, eh? That paragraph is so LLM it hurts, and probably a good chunk of your entire comment is too.
Well done! The very last paragraph is a patische from 5 different times I asked it to make a closing paragraph. Not even once did the actual output sound natural so I picked and chose different sentences until I got something that seemed better but yeah, each and every single word there came from an LLM. However I will say that just as Collage Art is considered Art by the Artist even though none of the pieces might be created by them, that last paragraph is still human because I did the curation and structuring.
Honestly I was hoping nobody would notice and then I'd spring it onto the unsuspecting populace of The Motte 3 days down the line...
The rest of the post is completely human generated by yours truly (artisanal tokens, so they say). If you think it's by Gemini 2.5 Pro I consider that to be a compliment as it's genuinely a better writer than I am. Failure to notice and remove the em dash is completely on me, ma faute.
No, this is not cute or clever.
We're still formulating exactly what our AI policy is, but we've certainly made it clear before that posting LLM output without declaring it to be so, especially as an attempt at a "gotcha," is low effort and not actual discourse. Consider this a formal warning, and we're likely to just start banning people who do this in the future.
I think a loooong effortpost should be allowed to have 1 paragraph of aislop as long as it's not relevant to the argument and can be deleted without hurting it. It would be a fun challenge for aihunters to find it. Maybe with a disclosure or something.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link