A lot of the reason why I even ask other people is not to get a simple technical answer but to connect, understand another person's unexepected thoughts, and maybe forge a collaboration –– in addition to getting an answer of course. Real people come up with so many side paths and thoughts, whereas AI feels lifeless and drab.
To me, someone pasting in an AI answer says: I don't care about any of that. Yeah, not a person I want to interact with.
I can buy into this. I always thought it was rude or at least insulting when Hollywood robotically creates slop movies. As in, of course they can do it, but damn is it insulting. There really are two types of people in the world:
a) Quantity > Quality if it prints $$$.
or
b) Quality > Quantity if it feels like the right thing to do.
Witnessing type A at scale is a first-class ticket into misanthropy.
I really wish some of my coworkers would stop using LLMs to write me emails or even Teams messages. It does feel extremely rude, to the point I don't even want to read them anymore.
I don't mind people using AI to help refine their thoughts and proof their output but when it is used in absence of their own thoughts I am starting to value that person a little bit less.
I find it as yet another way to externalize costs: I spend 0 time thinking, I dump AI slop on you and ask you to review it or refute me with the nonsense that I just sent you.
Last time someone did this to me I sent them a few other answers by the same LLM to the same prompt, all different, with no commentary.
Cause all an LLM is, is a reflection of its input.
Garbage in garbage out.
If we're going to have this rule about AI, maybe we should have it about... everything. From your mom's last Facebook post, to what is said by influencers to this post...
I got a feature request in the form of a PR a few months that said "chatgpt generated this as a possible implementation, does it work"?
I stopped there and replied that if you don't care enough to test if it works, then clearly you don't actually want the feature, and closed the ticket.
I have gotten other PRs that are more in the form of "hey I don't know what I'm doing. I used GPT but and it seems to work but I don't understand this part". I'm happy to help point in the right direction for those. Because an least they're trying. And seem like this is part of their learning.
... Or they just asked jippity to make it seem that way.
It gets interesting once you start a discussion about a topic with someone who had ChatGPT doing all the work. They often do not have the same in-depth understanding of what is written there vs. someone who wrote it themselves. Which may not come as a surprise, but yet - here we are. It‘s these kind of discussions I find exhausting, because they show no honesty and no interest by the person I'm interacting with. I usually end these conversations quickly.
Applications could automatically insert subtle icons next to messages that are automatically generated. It wouldn't work for copy-and-pasted text but it's a start.
LLMs are very very good at adding words in a way that looks "well written" (to our current mental filters) without adding meaning or value.
I wonder how long it will be before LLM-text trademarks become seen as a sign of bad writing or laziness instead? And then maybe we'll have an arms race of stylistic changes.
---
Completely agree with the author:
Earlier this week I asked Claude to summarize a bunch of code files since I was looking for a bug. It wrote paragraphs and had 3 suggestions. But when I read it, I realized it was mostly super generic and vague. The conditions that would be required to trigger the bug in those ways couldn't actually exist, but it put a lot of words around the ideas. I took longer to notice that they were incorrect suggestions as a result.
I told it "this won't happen those ways [because blah blah blah]" and it gave me the "you are correct!" compliment-dance and tried again. One new suggestion and a claimed reason about how one of its original suggestions might be right. The new suggestion seemed promising, but I wasn't entirely convinced. Tried again. It went back to the first three suggestions - the "here's why that won't happen" was still in the context window, but it hit some limit of its model. Like it was trying to reconcile being reinforcement-learning'd into "generate something that looks like a helpful answer" with "here is information in the context window saying the text I want to generate is wrong" and failing. We got into a loop.
It was a rare bug so we'll see if the useful-seeming suggestion was right or not but I don't know yet. Added some logging around it and some other stuff too.
The counterfactuals are hard to evaluate:
* would I have identified that potential change quicker without asking it? Or at all?
* would I have identified something else that it didn't point out?
* what if I hadn't noticed the problems with some other suggestions and spent a bunch of time chasing them?
The words:information ratio was a big problem in spotting the issues.
So was the "text completion" aspect of "if you're asking about a problem here, there must be a solution I can offer" RL-seeming aspect of its generated results. It didn't seem to be truly evaluating the code then deciding so much as saying "yes, I will definitely tell you there are things we can change, here are some that seem plausible."
Imagine if my coworker had asked me the question and I'd just copy-pasted Claude's first crap attempt to them in response? Rude as hell.
I have one of those coworkers. I tell him I have a problem with a missing BIOS setting. He comes back 2 minutes later "Yeah I asked an LLM and it said to go into [submenu that doesn't exist] and uncheck [setting I'm trying to find].
What's even more infuriating is that he won't take "I've checked and that submenu doesn't exist" for an answer and insists to check again. Had to step away for a fag a few times in fear of putting his face through the desk.
Yes it is absolutely rude in many contexts. In a team context you are looking for common understanding and being “on the same page”. If someone needs to consult AI to get up to speed that’s fine, then their interaction with you should reflect what they have learned.
I couldn't disagree more. Its like someone going to Wikipedia to helpfully copy and paste a summary of an issue. Fast and with a good enough level of accuracy.
Generally the AI summaries I see are more topical and accurate than the many other comments in the thread.
Someone telling you about a conversation they had with ChatGPT is the new telling someone about your dream last night (which sucks because I’ve had a lot of conversations I wanna share lol).
The author was thinking "boring and uninteresting" but settled on the word "rude." No, it's not rude. Emailing your co-workers provactive political memes or telling someone to die in a fire is rude. Using ChatGPT to write and being obvious about it marks you as an uninteresting person who may not know what they are talking about.
On the other hand, emailing your prompt and the result you got can be instructive to others learning how to use LLMs (aren't we all?) We may learn effective prompt techniques or decide to switch to that LLM because of the quality of the answer.
I recently had a non-technical person contest my opinion on a subtle technical issue with ChatGPT screenshots (free tier o4) attached in their email. The LLM wasn't even wrong, just that it had the answer wrapped in customary platitudes to the user and they are not equipped to understand the actual answer of the model.
The problem here is that I’ve been accused multiple times of using LLMs to write slop when it was genuinely written by myself.
So I apologized and began actually using LLMs while making sure the prompt included style guides and rules to avoid the tell tale signs of AI. Then some of these geniuses thanked me for being more genuine in my response.
A lot of this stuff is delusional. You only find it rude because you’re aware it’s written by AI. It’s the awareness itself that triggers it. In reality you can’t tell the difference.
On the whole it's considered bad to mislead people. If my love letter to you is in fact a pre-written form, "my darling [insert name here]", and you suspect, but your suspicion is just baseless paranoia and a lucky guess, I suppose you're being delusional and I'm not being rude. But I'm still doing something wrong. Even if you don't suspect, and I call off the scam, I was still messing with you.
But the definition of being "misleading" is tricky, because we have personas and need them in order to communicate, which in any context at all is a kind of honest, sincere play-acting.
60 comments
[ 0.18 ms ] story [ 56.3 ms ] threadTo me, someone pasting in an AI answer says: I don't care about any of that. Yeah, not a person I want to interact with.
a) Quantity > Quality if it prints $$$.
or
b) Quality > Quantity if it feels like the right thing to do.
Witnessing type A at scale is a first-class ticket into misanthropy.
Last time someone did this to me I sent them a few other answers by the same LLM to the same prompt, all different, with no commentary.
Cause all an LLM is, is a reflection of its input.
Garbage in garbage out.
If we're going to have this rule about AI, maybe we should have it about... everything. From your mom's last Facebook post, to what is said by influencers to this post...
Say less. Do more.
I stopped there and replied that if you don't care enough to test if it works, then clearly you don't actually want the feature, and closed the ticket.
I have gotten other PRs that are more in the form of "hey I don't know what I'm doing. I used GPT but and it seems to work but I don't understand this part". I'm happy to help point in the right direction for those. Because an least they're trying. And seem like this is part of their learning.
... Or they just asked jippity to make it seem that way.
That is not an excuse for it being poorly done or unvetted (which I think is the crux of the point), but it’s important to state any sources used.
If i don’t want to receive AI generated content, i can use the attribution to filter it out.
I wonder how long it will be before LLM-text trademarks become seen as a sign of bad writing or laziness instead? And then maybe we'll have an arms race of stylistic changes.
---
Completely agree with the author:
Earlier this week I asked Claude to summarize a bunch of code files since I was looking for a bug. It wrote paragraphs and had 3 suggestions. But when I read it, I realized it was mostly super generic and vague. The conditions that would be required to trigger the bug in those ways couldn't actually exist, but it put a lot of words around the ideas. I took longer to notice that they were incorrect suggestions as a result.
I told it "this won't happen those ways [because blah blah blah]" and it gave me the "you are correct!" compliment-dance and tried again. One new suggestion and a claimed reason about how one of its original suggestions might be right. The new suggestion seemed promising, but I wasn't entirely convinced. Tried again. It went back to the first three suggestions - the "here's why that won't happen" was still in the context window, but it hit some limit of its model. Like it was trying to reconcile being reinforcement-learning'd into "generate something that looks like a helpful answer" with "here is information in the context window saying the text I want to generate is wrong" and failing. We got into a loop.
It was a rare bug so we'll see if the useful-seeming suggestion was right or not but I don't know yet. Added some logging around it and some other stuff too.
The counterfactuals are hard to evaluate:
* would I have identified that potential change quicker without asking it? Or at all?
* would I have identified something else that it didn't point out?
* what if I hadn't noticed the problems with some other suggestions and spent a bunch of time chasing them?
The words:information ratio was a big problem in spotting the issues.
So was the "text completion" aspect of "if you're asking about a problem here, there must be a solution I can offer" RL-seeming aspect of its generated results. It didn't seem to be truly evaluating the code then deciding so much as saying "yes, I will definitely tell you there are things we can change, here are some that seem plausible."
Imagine if my coworker had asked me the question and I'd just copy-pasted Claude's first crap attempt to them in response? Rude as hell.
What's even more infuriating is that he won't take "I've checked and that submenu doesn't exist" for an answer and insists to check again. Had to step away for a fag a few times in fear of putting his face through the desk.
Generally the AI summaries I see are more topical and accurate than the many other comments in the thread.
On the other hand, emailing your prompt and the result you got can be instructive to others learning how to use LLMs (aren't we all?) We may learn effective prompt techniques or decide to switch to that LLM because of the quality of the answer.
So I apologized and began actually using LLMs while making sure the prompt included style guides and rules to avoid the tell tale signs of AI. Then some of these geniuses thanked me for being more genuine in my response.
A lot of this stuff is delusional. You only find it rude because you’re aware it’s written by AI. It’s the awareness itself that triggers it. In reality you can’t tell the difference.
This post, for example.
And then "echoborgs": https://en.wikipedia.org/wiki/Echoborg
On the whole it's considered bad to mislead people. If my love letter to you is in fact a pre-written form, "my darling [insert name here]", and you suspect, but your suspicion is just baseless paranoia and a lucky guess, I suppose you're being delusional and I'm not being rude. But I'm still doing something wrong. Even if you don't suspect, and I call off the scam, I was still messing with you.
But the definition of being "misleading" is tricky, because we have personas and need them in order to communicate, which in any context at all is a kind of honest, sincere play-acting.