gonzalohm
7 hours ago
It's not just the internet, it's everything.
Every document I read at work has been AI generated at some capacity. It's becoming so annoying that I have to use AI to create a summary.
I don't know if anyone agrees, but AI generated documents are so verbose and explain things in an overcomplicated way that makes me exhausted after reading. Sometimes I don't even understand what AI is trying to explain
elevation
7 hours ago
> in an overcomplicated way that makes me exhausted after reading
This. I can't even listen to youtube videos any more because of their chat-gpt generated scripts. LLMs will readily generate several pages of superfluous prose, rehashing the same idea with punchy witticisms for no reason. But it uses such compelling connecting words that you might be tricked into believing there will be a payoff long past the point of diminishing returns.
I have been enjoying reading literary classics for their poetic concision. Sometimes, a bit of context can allow a simple sentence to form a picture in your mind's eye that would take a few minutes to explain syllogistically. I wish to communicate more like that.
phishin
7 hours ago
Your comment’s load-bearing property, asserts on the geometry that states it.
notpachet
7 hours ago
You've hit on something important, and it's real, not cosmetic.
genxy
7 hours ago
I know you guys are joking, but this is like peeing in the sauna or shitting upwind. These serious writing samples will just enfreshen the training data.
glouwbug
7 hours ago
This was really well written, and you hit on a serious point that most miss
mckee_plus_plus
7 hours ago
Claude's butthole logo, meet Idiocracy
ahhssja
7 hours ago
[dead]
CrzyLngPwd
7 hours ago
A little sick in the throat is normal.
ahhssja
6 hours ago
[dead]
gofreddygo
7 hours ago
No kidding. I got a feature handed down (with links to 4 other markdown docs) for what was discussed in person as "api to record audit events". The result is <goes to AI for word count then gives up>
wc .\ledger_reqs.md
1090 9347 68071 .\ledger_reqs.mdOptionOfT
6 hours ago
For too long we have used length of document as a measure of quality.
user
7 hours ago
icedchai
7 hours ago
It's happening everywhere. We have people using AI to generate documents, then using the same AI to read/summarize them. What is the point? The writing itself, especially with recent Claude models, is often incredibly awkward.
spicyjpeg
7 hours ago
I like to call this phenomenon "reverse compression", or occasionally "depression". It's even lossy and subject to generational loss.
icedchai
7 hours ago
Hah. "Depression" fits well for a few reasons.
cameronh90
7 hours ago
The logical answer is to send what you would have put into AI.
However, terse lists of bullets without any padding or phatic expressions are considered rude. Perhaps previously a long message indicated you'd put time and thought into it.
Maybe AI will invert the social signalling, and make conciseness polite. As it should be.
gonzalohm
6 hours ago
It's the wet dream of any CEO. Subscription model for companies to pay to do tasks and subscription model for companies to understand the output of the subscription model. Amazing, they should do another Silicon valley season
pixl97
6 hours ago
>What is the point?
What is the point is a question to a whole lot of human practices outside of AI, well, that I should say that AI is pointing out.
Huge numbers of us have mostly useless jobs. That's not to say they are completely useless, but maybe like at most 50% of the job is useful. The rest just gets filled with busy work and worthless text generation and TPS reports.
icedchai
5 hours ago
Agreed, a ton of white-collar job activity is useless: writing reports nobody wants to write for people who don't actually want to read them anyway. Mass AI-induced unemployment is a real possibility within 5 to 10 years, maybe sooner.
jstummbillig
7 hours ago
Most likely outcome seems to be that this period will be over very quickly, and we will have better writing than ever before, forever.
We should probably put some more thought toward that part and what it will mean.
nxc18
7 hours ago
Why do you think that? We’ve had this exact problem since the GPT3 moment and it has not improved at all. Not even a little, unlike agentic coding. Why do we expect a horizontal line to suddenly go exponential?
Also, it seems that writing encodes far more information about personality and mental state of the writer than it seems at first glance. A key problem is that machine-generated writing does not have this, it is obvious that it doesn’t have this, and there is no path towards fixing it.
jstummbillig
7 hours ago
What exact problem? Verbosity? Quality?
gonzalohm
6 hours ago
The quality is eroded by the verbosity
ahhssja
6 hours ago
[dead]
gizmo686
7 hours ago
Why is that the most likely outcome?
The history of automation is filled with cases where the result has been lower quality goods at lower prices.
pixl97
7 hours ago
Yea, this is definitely a place where the positive future outlooks have to state why.
The cost for text generation will trend toward zero.
The amount of models that will be able to generate text will keep expanding making what text is generated or not will become more difficult.
The way actual humans write will change because of influence by AI mushing up what is human and what is not human further increasing the impossibility of determining what is AI and what is not.
Freedom of speech laws in the US will shoot down any broad legal requirements to mark text as AI generated or not.
The consumer financial incentives to go to the 'better' and more expensive products are questionable at best. People consuming a lot of shitty shit because it's cheap is a pretty standard human behavior. It doesn't mean good products won't exist. It means good products will have to do a shit load of work to stand above and beyond the slop, and with systems like the US, these better products are apt to be bought by VC to exploit as much money out of the product by turning it into slop and getting rid of the human assets behind it.
Also..
You only notice 'slop', not necessarily things that are written by AI (or lazy humans).
DrewADesign
7 hours ago
> Why is that the most likely outcome?
Listen, very smart people are very enthusiastic about other very smart people using AI to help AI learn to fix AI. Pretty much open-and-shut, IMO.
rwmj
7 hours ago
Those people will stop really understanding anything because they defer all their writing (and so thinking) to machines. They will get stupider and stupider.
odie5533
7 hours ago
We're coming up on 4 years into the AI-slop season of the internet. Just one more year and then we all go extinct?
threethirtytwo
7 hours ago
This can be improved. As AI grows it will be better at explaining. For sure people working on frontier models have identified this as a problem.
owebmaster
7 hours ago
They didn't or they can't fix. Claude opus 5 writing is much worse than 4 years older models like gpt4
surcap526
7 hours ago
[dead]
marknutter
7 hours ago
It's still very early days. I have no doubt that etiquette and norms will be established that alleviate these issues.