It’s either from the context or a deliberately chosen chat style, I presume.
One thing that isn’t emphasized enough about the ChatGPT appis that is very “far” from the LLM. There’s a ton of stuff wrapping around it and getting injected into the context, tons of tracking and personalization and engagement maximization.
And the Chat models are finetunes to oblivion. They are deep fried.
It’s definitely deliberately chosen. There’s no way it’d randomly just go into Reddit speak; that’s not how an LLM like this works: it maintains that generic, stale toast tone unless you ask it otherwise.
Which also makes me question what the prompts before this were and makes me assume this is faked. Like “pretend to be shocked” faked.
There’s no way it’d randomly just go into Reddit speak; that’s not how an LLM like this works: it maintains that generic, stale toast tone unless you ask it otherwise.
An LLM will respond in whatever way it’s been tuned to, if Sam Altman wants it to sound like a redditor, it’ll sound like a redditor.
It’s like how Grok responds in the most cringe unfunny way possible, it’s been finetuned to respond like Elon Musk.
The mistake here is thinking it would “randomly” go into Reddit speak. There’s a lot random about AI but not all of it is, and both it and the systems built on top of it are far more sophisticated than you’re giving them credit for.
The prompts and context being invisibly sent by the app to shape its interpretation and its response are exactly what “ask it otherwise”, and unless you know exactly what you’re doing (and probably are using an actual open source, open principles frontend) you’ll never know exactly what your tooling is actually telling the LLM itself. That’s obfuscated behind layers of actual software interpreting and analyzing everything you type. We’ve had software doing stuff like that since before LLMs were a thing and it’s only gotten more advanced now that everyone on the planet is using “chat” to talk to LLMs. If you think what you type into the chat goes directly into an LLM that couldn’t be further from the truth. It’s massaged and decorated and marked up and embedded into a whole data package the LLM gets provided behind the scenes, and even then there are additional shaping layers built into the model itself to further transform what it’s receiving. There’s no guarantee you’re even using the exact same model as someone else, even if they both say ‘GPT-5’ there’s no proof and no reason to believe you’re not potentially getting ‘GPT-5-Lemmyuser’ or even ‘GPT-5-TheTechnician27’ behind the scenes. Even if it’s only GPT-5 with a few extra layers tacked on, how would you know the difference… and they’re certainly not going to tell you.
You launched into an entire spiel about “randomly” when I clearly meant it in the colloquial sense of “out of nowhere” – that in 2026, it’s not going to inexplicably break tone so heavily.
[Input is] massaged and decorated and marked up and embedded
I understand how tokenization etc. works. I’ve taken multiple machine learning courses. That has nothing to do with the fact that, again, “I’d call the police 😂” ain’t happening unless you intentionally pull it away from its preset tone. Having not used ChatGPT etc. for myself but seeing thousands of other people use it, I feel pretty confident in that.
This is especially true because the RAM crisis has been going on for a long enough time now that this response is clearly faked; this isn’t new information a model like ChatGPT could be “surprised” by (using “surprised” in scare quotes because now I’m afraid you’ll launch into another spiel if I don’t disclaim that ChatGPT isn’t literally capable of surprise in the sense of an emotion).
Also, please break up your paragraphs. I’m bad about this sometimes too.
It’s not a pure LLM though. It’s ChatGPT. That’s what I’m saying; users have no idea what else is in the context, and it could easily be something that pushes it to “Reddit speak.”
When I’ve seen my parents use it, it brings up things it could not possibly know from the current conversation, so its definitely inserting previous context at the very least.
When I’ve seen my parents use it, it brings up things it could not possibly know from the current conversation, so its definitely inserting previous context at the very least.
I have literally no idea what you’re so confidently talking about by “pure LLM”. It’s hooked into other stuff too that lets it e.g. properly do math (I’d assume it can by this point?), generate/analyze images, etc. But that has fuck-all to do with whether it can bring previous context into a conversation. Go read the 2017 paper “Attention is All You Need”; literally the whole point of a transformer model is that it can assess the surrounding context of input tokens out to a basically arbitrary depth thanks to its heavy parallelization. It’s obviously imperfect; go watch the most recent Kitboga video (clickbait, sorry) where he creates Gertrude the Pig by leading a scambot away from its training data.
In fact, older LLMs (similarly “pure”, again whatever that means) had far worse problems with being cursed by memory. That’s why the LSTM was invented from the RNN: because the model needed to have a way to forget accrued past context over time, or the results would quickly corrupt on longer inputs. (Keep in mind that with LSTMs, I’m mainly talking about performing more basic things like sentiment analysis.) That didn’t make it “impure”; it just made it not as good at its job as a GPT.
Anyway, unless previously prompted to change to this tone: no, in 2026, it’s not just going to start saying “I’d call the police 😂” in response to high component prices.
The ChatGPT app injects history and all sorts of stuff into context automatically, I’ve seen it do it, and that might include something (like the user’s previous chats) that unintentionally give a style for the model to adopt.
To be specific, I saw ChatGPT reference a specific appointment of a family member in a context that was absent of that information, on a topic that had nothing to do with it. The only way it could have possibly known that was injecting the previous chat into context, silently.
That’s all I’m saying.
LLM architecture has nothing to do with that; its being done outside the LLM. By “pure”, I meant a chat interface where one can see the full prompt as input.
What I’m saying is, we don’t know if that’s the cause or not until we see the context. But it could plausibly be unintentional on the user’s part.
I promise I was not being “hostile”. I led with “I have literally no idea what you’re talking about”, and that’s because I frankly did not. I appreciate you clarifying, although I think dividing it into “pure” and (consequently) “impure” isn’t sensible terminology. An LLM is just that: a model trained on a huge text corpus.
I still get what you’re going for now, namely that output can be “tainted” by previous input, hence “purity”. I think something like “isolated input” would be more apt, although that’s also ad hoc, and there’s probably real, established terminology out there somewhere (that maybe I’m even forgetting).
It’s either from the context or a deliberately chosen chat style, I presume.
One thing that isn’t emphasized enough about the ChatGPT appis that is very “far” from the LLM. There’s a ton of stuff wrapping around it and getting injected into the context, tons of tracking and personalization and engagement maximization.
And the Chat models are finetunes to oblivion. They are deep fried.
It gives me creepy Facebook vibes.
It’s definitely deliberately chosen. There’s no way it’d randomly just go into Reddit speak; that’s not how an LLM like this works: it maintains that generic, stale toast tone unless you ask it otherwise.
Which also makes me question what the prompts before this were and makes me assume this is faked. Like “pretend to be shocked” faked.
You guys are working with 5 year old info on llms now. Sounding like grandma over here “what’s this technology you kids are using…”
An LLM will respond in whatever way it’s been tuned to, if Sam Altman wants it to sound like a redditor, it’ll sound like a redditor.
It’s like how Grok responds in the most cringe unfunny way possible, it’s been finetuned to respond like Elon Musk.
The mistake here is thinking it would “randomly” go into Reddit speak. There’s a lot random about AI but not all of it is, and both it and the systems built on top of it are far more sophisticated than you’re giving them credit for.
The prompts and context being invisibly sent by the app to shape its interpretation and its response are exactly what “ask it otherwise”, and unless you know exactly what you’re doing (and probably are using an actual open source, open principles frontend) you’ll never know exactly what your tooling is actually telling the LLM itself. That’s obfuscated behind layers of actual software interpreting and analyzing everything you type. We’ve had software doing stuff like that since before LLMs were a thing and it’s only gotten more advanced now that everyone on the planet is using “chat” to talk to LLMs. If you think what you type into the chat goes directly into an LLM that couldn’t be further from the truth. It’s massaged and decorated and marked up and embedded into a whole data package the LLM gets provided behind the scenes, and even then there are additional shaping layers built into the model itself to further transform what it’s receiving. There’s no guarantee you’re even using the exact same model as someone else, even if they both say ‘GPT-5’ there’s no proof and no reason to believe you’re not potentially getting ‘GPT-5-Lemmyuser’ or even ‘GPT-5-TheTechnician27’ behind the scenes. Even if it’s only GPT-5 with a few extra layers tacked on, how would you know the difference… and they’re certainly not going to tell you.
You launched into an entire spiel about “randomly” when I clearly meant it in the colloquial sense of “out of nowhere” – that in 2026, it’s not going to inexplicably break tone so heavily.
I understand how tokenization etc. works. I’ve taken multiple machine learning courses. That has nothing to do with the fact that, again, “I’d call the police 😂” ain’t happening unless you intentionally pull it away from its preset tone. Having not used ChatGPT etc. for myself but seeing thousands of other people use it, I feel pretty confident in that.
This is especially true because the RAM crisis has been going on for a long enough time now that this response is clearly faked; this isn’t new information a model like ChatGPT could be “surprised” by (using “surprised” in scare quotes because now I’m afraid you’ll launch into another spiel if I don’t disclaim that ChatGPT isn’t literally capable of surprise in the sense of an emotion).
Also, please break up your paragraphs. I’m bad about this sometimes too.
It’s not a pure LLM though. It’s ChatGPT. That’s what I’m saying; users have no idea what else is in the context, and it could easily be something that pushes it to “Reddit speak.”
When I’ve seen my parents use it, it brings up things it could not possibly know from the current conversation, so its definitely inserting previous context at the very least.
I have literally no idea what you’re so confidently talking about by “pure LLM”. It’s hooked into other stuff too that lets it e.g. properly do math (I’d assume it can by this point?), generate/analyze images, etc. But that has fuck-all to do with whether it can bring previous context into a conversation. Go read the 2017 paper “Attention is All You Need”; literally the whole point of a transformer model is that it can assess the surrounding context of input tokens out to a basically arbitrary depth thanks to its heavy parallelization. It’s obviously imperfect; go watch the most recent Kitboga video (clickbait, sorry) where he creates Gertrude the Pig by leading a scambot away from its training data.
In fact, older LLMs (similarly “pure”, again whatever that means) had far worse problems with being cursed by memory. That’s why the LSTM was invented from the RNN: because the model needed to have a way to forget accrued past context over time, or the results would quickly corrupt on longer inputs. (Keep in mind that with LSTMs, I’m mainly talking about performing more basic things like sentiment analysis.) That didn’t make it “impure”; it just made it not as good at its job as a GPT.
Anyway, unless previously prompted to change to this tone: no, in 2026, it’s not just going to start saying “I’d call the police 😂” in response to high component prices.
Whoa. What’s with the jump to being so hostile?
The ChatGPT app injects history and all sorts of stuff into context automatically, I’ve seen it do it, and that might include something (like the user’s previous chats) that unintentionally give a style for the model to adopt.
To be specific, I saw ChatGPT reference a specific appointment of a family member in a context that was absent of that information, on a topic that had nothing to do with it. The only way it could have possibly known that was injecting the previous chat into context, silently.
That’s all I’m saying.
LLM architecture has nothing to do with that; its being done outside the LLM. By “pure”, I meant a chat interface where one can see the full prompt as input.
What I’m saying is, we don’t know if that’s the cause or not until we see the context. But it could plausibly be unintentional on the user’s part.
I promise I was not being “hostile”. I led with “I have literally no idea what you’re talking about”, and that’s because I frankly did not. I appreciate you clarifying, although I think dividing it into “pure” and (consequently) “impure” isn’t sensible terminology. An LLM is just that: a model trained on a huge text corpus.
I still get what you’re going for now, namely that output can be “tainted” by previous input, hence “purity”. I think something like “isolated input” would be more apt, although that’s also ad hoc, and there’s probably real, established terminology out there somewhere (that maybe I’m even forgetting).