On OpenRouter there are 11 third-party inference providers hosting DeepSeek V4 Pro right now, of which 8 are US-based and 7 of those have zero data-retention policies (which I mention to rule out any claims of "oh they're making up the money by logging all your data"). This is a 1.6T-A49B model, so a bit bigger than Sonnet (~2/3rds the size) and a bit smaller than Opus (~3x as large). These third parties are almost perfectly interchangeable via OpenRouter as a marketplace, so they have no incentive to offer any sort of "growth pricing" below costs, and they serve it at $3.48/Mtok out.
Kimi K2.6 is 1T-A32B with a slightly less computationally efficient architecture, and is served at around $3.50/Mtok out by 9 US ZDR providers.
Unless you think that either the generally accepted size estimates for Anthropic/OpenAI models are wildly off or those companies are a lot worse at serving models efficiently, Anthropic and OpenAI are probably making around 5-8x margins on their API costs.
The cost of training new models is of course a major factor not counted here. Depending on how you want to think about that this may or may not make them net profitable. I remember one of those CEOs gave an interview a while back where they described it as a series of independent investments, where each model they train is net-positive in revenue by EOL just from its own inference, but I don't know whether that's still true or not.
Regardless, the point is that if they stopped training new models today, both Anthropic and OpenAI are making incredibly generous profits on their API inference.
This phenomenon (and/or the fear of being falsely accused) is why I find myself being an annoying unpaid evangelist for Pangram every few days.
Right now a lot of people are developing their own heuristics, some much more well-calibrated than others, for sniffing out AI generated text but it's kind of an unfalsifiable accusation since you're really just saying "this sounds like slop, I can tell from the words and from seeing lots of slop in my time".
Linking to an AI writing detector serves as a somewhat objective second opinion, but half the time if I say "Pangram confirmed it" I'll see at least one person pulling the old "yeah but GPTZero is crap so that means AI writing detection can't work in principle" maneuver, which strikes me a lot like someone who tried ChatGPT in 2023, decided AI was all hype, and never re-evaluated that stance.
It would be really helpful for the discourse if we could get to a point where people generally accept one or more _actually-good_ LLM writing detectors as a reliable tool such that a "this is AI" judgement from them is accepted as reasonable proof of guilt, solely so that we can then extend the benefit of the doubt to anybody else.
You are incorrect. There, now we've both made unsupported assertions. Care to provide any evidence for your position?
For what it's worth, when I provide a Pangram link it's because I can already tell something is AI and I'm attempting to provide objective third-party confirmation so the conversation doesn't just degrade into me asserting that I have superior taste to you.
AFAICT they're basically saying "hey lawmakers we're not going to rock the boat or take a stand so in exchange pretty please don't lump us into the app store category"
Have courage and trust your own instincts. Unless one is extremely disagreeable it's very tempting to hedge and avoid outright saying "this is AI" just in case you're wrong, but if you're literate and regularly exposed to AI outputs your instincts are likely quite accurate.
In this particular case the linked article is definitely AI generated.
Indeed, consider these two posts linked below also from this blog. They look the same, they maintain the same impersonal writing style. There's no humanity to it at all.
They maintain such a consistent paragraph length that they're either a professional copyeditor or, as is clearly the case, are an LLM.
Humans deviate a lot more than this, they use run on sentences or lose the thread in their writing.
This blog however reads like every-other post on LinkedIn. Semi-professional tone, with a strong "You, Me" hook to most posts.
I encourage everyone to make an LLM-generated blog, don't post the articles anywhere, but generate one, to get a feeling for how these things write.
Because this is unmistakably LLM. I'd even go so far as to identify the model of these particular posts as ChatGPT.
Yet when we point this out, we're told it is "unmistakably human" and that we're rude for pointing it out.
What does that have to do with anything? These days any piece of text may or may not be AI generated (my money would be heavily on "no" for the post you asked about), but either way it isn't blatant slop so we can't tell.
It feels like you're trying for a lazy gotcha, but the actual point here is something like "AI models often generate writing with specific noticeable characteristics that make it obviously AI output, and TFA is an instance of such writing, and this should be called out when possible"
OTOH I’ve had blog posts I wrote two decades ago vehemently called out as AI generated. AI generated style unfortunately means writing that tested positively in human A/B testing, now over represented in a style used largely by AI.
So if you write in a way that engages the reader, you’re going to struggle not to use em dashes and the occasional a/b contrast, because those are challenging the reader to engage… but when overused, they not only don’t have the intended effect ( to break the reader out of passivity) , they also constitute a new kind of sin.
So no, don’t “trust your gut”. Trust the math. Is it too much? Or is it just trying to jar you out of not engaging with the prose?
But yeah, I’d say this article is likely written primarily with AI. Which doesn’t mean it’s not guided with intention and potentially important, it just means the article was probably commissioned and edited by a human, not written by one.
It's kind of funny when I open some books nowadays and the writing style and formatting just immediately scream LLM sometimes. Not because the book was AI-generated, most are too old, but because LLMs were simply trained on these exact books and are now reproducing their style, which I guess was either popular or selected during training.
Anyways, really hard to push through and I need to remind myself to judge the text by its meaning. But if it's some random blog, my "tolerance" is lower and I don't want to spend my time reading nonsense, I just can't stand the writing style anymore either.
> OTOH I’ve had blog posts I wrote two decades ago vehemently called out as AI generated. AI generated style unfortunately means writing that tested positively in human A/B testing, now over represented in a style used largely by AI.
Everytime I see this claim, I ask for links to those blog posts. I have yet to get any links to the so-called "human" pattern that AI uses.
This blog of my idle musings, specifically, has been a source of call-outs. In articles from back in 2013 of all things. I also noticed that( ChatGPT?) seems to have replied to one of my latest (2023) posts, which I find odd and improbable
I get what you mean though, to me I don’t see the hallmarks of AI writing, but you will find the occasional em-dash and contrasted-constructions. I think some people see an em-dash and decide them and there that it’s AI generated, probably because they are illiterate by any reasonable measure of the term.
I started off hedging but by the end of the comment came to think that AI use—or lack thereof—was actually beside the point. I have feelings with regards to the situation where “the situation” includes some largely irrelevant-to-writing things like the mainframization and the “feelings” are not nearly coherent enough to graduate to thoughts. Thus (unlike some others) I don’t think that calling out writers or warning readers about AI is all that useful (or for that matter courageous). With respect to writers who use AI due to a lack of confidence, it’s probably even harmful. (Saying that as a person who manages to absolutely suck in embarrassing ways in multiple foreign languages. And also in English but less obviously. And likely in my native language too due to lack of use.) Meanwhile, TFA makes a decent point, and I am in no position to criticize people for being wordy.
The thing is, by now it doesn’t actually matter if AI or not AI or partly AI or whatever, because the record scratch is still there and still breaks my immersion. I could be oversensitive (I definitely am to some other English-language things, and also feel that others are to yet other things like em dashes), but it feels like there’s a new language/social-signalling thing now, and you may have to avoid it even if you’re not an LLM.
> if you're literate and regularly exposed to AI outputs your instincts are likely quite accurate.
There's no basis for that. The reason experts - for example, scientists in their own field - use objective fact is because reasoning like the parent is highly unreliable. What evidence shows is that people way overestimate their own intuition. It's not 'courage', it's foolishness.
With the amount of AI-generated slop content on the front of HN these days, I'm honestly reconsidering visiting this site in the first place. What's the point? It seems better to curate RSS from existing known-good sources.
The art of essay-writing seems to not be something people here care about any more. If a human didn't bother to write it, why should I bother to read it?! Just post up the bullet points you would feed the LLM, and let the people who want to do so, post it into their own LLMs so they can make the Content and shovel it into their eyeballs by themselves, instead.
The Kobo Clara doesn't fit in my pocket while the X4 does. An e-reader that I actually have on-hand beats one that I leave in a drawer at home so it was an instant purchase for me the moment I saw one.
Tokens _are_ as cheap as keystrokes. A single keypress by a full-time SWE averages out to $0.005-$0.02 (depending on typing speed and TC). The relationship is obscured because the keystrokes are usually part of a fixed-price subscription plan but they absolutely have a cost. Prior to AI this was in fact a large reason everyone pontificated about concise programming languages and elegantly factoring problems and DRY and...
I can set up a drinking-bird-style device to press keys on a keyboard. My per-year operational cost is the same whether or not the bird presses keys.
Programmer time costs money. Nearly all of a programmer's time is spent understanding the system being maintained, the reasons for its existence, and the reasons for the proposed changes to it. If you're honestly such a fool as to think that DRY, "elegant" [0] design, and related things are about reducing the number of keys pressed rather than the time spent understanding the system and how to change it, then... well, I think you make an awfully good "AI Booster".
[0] Read as "easy to reason about, once you understand the core concepts"
The only correct way to ask an AI "why did you do that?" is in the sense of a blameless postmortem. You're the person responsible for giving the LLM appropriate context and instructions and guardrails, so the only reason you should ever ask a question like that is when you're genuinely trying to figure out how to improve those for next time. Every time I see people posting this sort of "apology" from an LLM it makes me cringe, feels only half a step away from outright AI psychosis.
That seems like unnecessary pedantry, only one step short of the common "everything is chemicals" retort. Egg yolks are an ingredient that contributes to the flavor and nutrient profile of the food while also emulsifying. In this context "emulsifiers" is presumably meant to refer to other ingredients added solely to modify the texture.
I disagree, I think “emulsifiers” is meant to be used as a scary long word that people don’t know what it means, and don’t know that mustard, egg yolk and more are common emulsifiers. And it underlies the pseudoscience at the core of what is more marketing than real health.
TL;DR: Devs didn't know what they were doing and turned off autovacuum and eventually it broke, then the author decided to have an AI slop out an article about the incident which may or may not have actually occurred.
Kimi K2.6 is 1T-A32B with a slightly less computationally efficient architecture, and is served at around $3.50/Mtok out by 9 US ZDR providers.
Unless you think that either the generally accepted size estimates for Anthropic/OpenAI models are wildly off or those companies are a lot worse at serving models efficiently, Anthropic and OpenAI are probably making around 5-8x margins on their API costs.
The cost of training new models is of course a major factor not counted here. Depending on how you want to think about that this may or may not make them net profitable. I remember one of those CEOs gave an interview a while back where they described it as a series of independent investments, where each model they train is net-positive in revenue by EOL just from its own inference, but I don't know whether that's still true or not.
Regardless, the point is that if they stopped training new models today, both Anthropic and OpenAI are making incredibly generous profits on their API inference.