Have you ever read someone else's conversation with an LLM?
759 karma · joined September 3, 2025
I think that Google initially came up with transformer architecture to use it for translation, so...
At least that was the case before LLMs became a thing, now I'm not sure anymore.
Created confusion and frustration will make it much harder to separate signal from the noise for most people.
And I did not speak out
Because I was not using em dashes
Then they claimed that if you're crammar is to gud you r not hmuan
And I did not spek aut
Because mi gramar sukcs
Then they claimed that if you actually read the article that you are trying to discuss you are not human...
As far a I can understand the study design, it can establish a correlation but bot a causal relationship.
So, instead of implying that using computer up to "2.4 h/day" decreases risk of dementia, but higher screen time increases it, we could as well conclude that dementia patients are either unable to use a PC (< 2.4 h/day) or unable to stop using a PC (> 2.4 h/day).
I'm not confident at all at my ability to interpret this study, since it is definitely not my area of expertise, so, please, correct me if I'm wrong.
Update #1, after reading the article more carefully:
> "First, as an observational cohort study, causal relationships cannot be inferred; observed associations may reflect residual or unmeasured confounding and reverse causation."
that's what I said
> Second, a single baseline self-report likely introduced measurement error, missed secular shifts in digital-media use, and lacked content/context; session fragmentation, breaks, concurrent media, timing (e.g., evening vs daytime), and device type were unavailable.
How much do we really trust a dementia patient's self-report on device usage? I wouldn't even trust a healthy person's self-report. I think that anyone who ever tried to track screen time on computer or on the phone can understand how unreliable are your intuitive guesstimates compared to actual measurements.
Had to turn off the "Night Light" (reducing blue) and set brightness to max.
If you are unable to run the multimillion training, then any kind of security audit of the training code is absolutely meaningless, because you have no way to verify that the weights were actually produced by this code.
Also, the analogy with source code/binary code fails really fast, considering that model training process is non-deterministic, so even if are able to run the training, then you get different weights than those that were released by the model developers, then... then what?
These days you might even vibecode it in one prompt, although coding it manually is not super hard either.
def replace_em_dash(text: str) -> str:
"""
+-------------------+
| ( ͡° ͜ʖ ͡° ) |
+-------------------+
"""
return text.replace("—", "\U00010EAD\U00010EAC")I didn't expect ChatGPT to make such trivial mistake, although, I have no idea which model do they use on the free plan these days.
The correct code is, of course:
text.replace("—", "\U00010EAD\U00010EAC")
def replace_em_dash(text: str) -> str:
"""
+-------------------+
| ( ͡° ͜ʖ ͡° ) |
+-------------------+
"""
return text.replace("—", "\u10EAD\u10EAC")
[citation needed]
> meaningful review No that I think about it, maybe for the first time in history it's actually feasible to review all the code in the repos using LLMs. Before LLMs were a thing, for any big project that would be way too much work to realistically do it.
Also, someone can provide code review of publicly available dependencies as a service, to avoid wasting tokens of reviewing same code again and again by each dev locally on their machine.
U wonder if anyone is already working on such service...
If the discussion is related to a public project, like in the examples in this comment: https://news.ycombinator.com/item?id=47303604
...you can use existing communication channels (website, readme on github) to ask people for an invite to participate in it.
New account can be invited or vouched for by an old account with good karma. If an account that you vouched for starts spamming and/or slopposting, you lose your vouching for abilities for a period of time or forever.
Is it really that big of an issue if your package manager pins dependencies by hash?
I guess, public package registry can be down an brake your pipeline, that's a risk. But I don't see how it introduces any new security problems.
> I want to do good engineering, not produce slop, but for 1 min of prompting, 5 mins of tidying, and 30 mins of review, we might save 2 days of eng time.
I don't really understand where do "2 days of engineering time" come from.
What exactly would prevent someone who does know the codebase do "1 min of prompting, 5 mins of tidying, and 30 mins of review" but then actually understand if changes make sense or not?
More general question: why do so many slopposters act like they are the only ones who have access to a genAI tool? Trust me, I also have access to all this stuff, so if I wanted to read a bunch of LLM-slop I could easily go and prompt it myself, there is no need to send it to me.
Related link: https://claytonwramsey.com/blog/prompt/ (hn discussion: https://news.ycombinator.com/item?id=43888803 )
> By exploiting vulnerabilities in the signaling layer, they could compromise a target’s client without their knowledge.
If the client is compromised, anything is possible.
[0] https://en.wikipedia.org/wiki/Wikipedia:No_original_research...
Like, "we don't like copyright, but since you insist on enforcing it and we can't do anything against it, we will invent a clever way to use your own rules against you".
I personally prefer light themes everywhere, both in IDEs and in the terminal. I thought that just choosing my own color scheme for 0-15 would give me the color pallette that I prefer, but because app developers like you for some reason decided that you know better what colors do I prefer, this is actually not enough. I also have to configure each TUI application separately to have the color scheme that I like.
And I do not understand why people do it. Like, why would you deliberately break the universal customization system and force users to use your own, specific to your app?
Honesty, each time I encounter an app that uses 16-255 colors, I feel like someone just violated my personal space and intruded into my chosen color pallette with their own colors that don't fit.