marvlouslie
kiwifarms.net
- Joined
- Nov 27, 2021
Most of those kinds of comments are almost always all botted. It's best just to ignore the AI. Dead Internet Theory is kinda real once you realize how much of the web is just ads and botted accounts.
Modern Web Woes - I'm mad at the internet
- Thread starter byuu
- Start date
-
Want to keep track of this thread?
Accounts can bookmark posts, watch threads for updates, and jump back to where you stopped reading.
Create account
Most of those kinds of comments are almost always all botted. It's best just to ignore the AI. Dead Internet Theory is kinda real once you realize how much of the web is just ads and botted accounts.
August Levasseur
kiwifarms.net
- Joined
- Dec 13, 2022
Ladybird can’t come soon enough
- Joined
- Dec 7, 2020
Human brain interfaces don't work that well IRL. The monkeys they test it on usually die from the installation procedure. It's like this old episode of Stargate SG-1 where a town is slowly killed off by an AI controlling the human brain interfaces. Whatever happens in the future, this vaporware product is trying to kill you.
- Joined
- Jul 2, 2025
YouTube has their priorities straight. -- This happens when you like a video that's considered scary/spooky/horror.
- Joined
- Jul 8, 2024
I hate it when the 2FA code field auto-submits after typing the last digit. There were multiple times when I mistyped the last digit and had to redo it all over.
- Joined
- Dec 7, 2020
Goodbye writing.
Goodbye thought.
Goodbye honesty.
Goodbye sincerity.
Goodbye love.
Goodbye meaning.
Goodbye ideas.
Goodbye self expression.
Goodbye world.
A poojeet in-between wiping his ass in a shitting field is going to text your email for you on his shit covered smart phone.
Goodbye thought.
Goodbye honesty.
Goodbye sincerity.
Goodbye love.
Goodbye meaning.
Goodbye ideas.
Goodbye self expression.
Goodbye world.
A poojeet in-between wiping his ass in a shitting field is going to text your email for you on his shit covered smart phone.
Last edited:
- Joined
- Oct 2, 2017
We copy-paste stock phrases from a word document to our emails on repeat all day. We've got a tool to do it directly inside outlook but that's apparently worthy of an entire conversation with superiors and what not. Honestly, if they told me to use this type of shit, I'd be thankful. It's corpo speak, who gives a shit.
- Joined
- Dec 7, 2020
No one is going to read it if they know an AI wrote it. There is no point in trying to fluff what you are saying with AI. It will become an cargo cult of overly wordy responses which say nothing and nobody reads.
Canned responses are not the same because there is a difference in writing a response to an email and then reusing it when different people write the same email to you multiple times than having an AI make up a response. You don't even know if the AI is going to do a good job. It would be very embarrassing to have an AI make promises or create impressions with the people you are emailing with that then become expectations put on you, which you were never aware of because you didn't grade the AI's email response before sending it because that would defeat saving time by using AI.
All these companies always make out that their products always work perfectly everytime. That’s not realistic, it’s only marketing. What Google says it will be like is not automatically what it will really be like. We all know what hell it was when Google tried to save their failing social media platform, Google+, by integrating it with YouTube, and how much it made the user experience on YouTube worse.
AbendCode12
kiwifarms.net
- Joined
- May 31, 2020
Then why are all those people paying subscriptions to create text and images? I think you rather mean he'd lose face if it was discovered he was outsourcing his daily work correspondence to a python script written by a jeet.
- Joined
- Jul 2, 2025
I'll just be sad that I won't have Saar Ismael Fujaheer Pageuin typing out personal emails to me anymore. It's my favorite form of entertainment out there, now even Jeets won't have a job anymore. Unless Gemini is the other AI, Actual Indians.
- Joined
- Nov 26, 2018
Who cares if AI writes everything, do you not re-read what you wrote and edit it a few times before hitting send?
- Joined
- May 4, 2023
Funny story about the uselessness of this. A client of mine tried and failed to have an LLM write his emails to me several times, and every time it was complete gibberish. Making up corpo-buzzwords on the spot to explain what he wanted to say. After billing him handsomely for my time he gave up and started using his words like a big boy.
To give it a chance, I tried it myself, asking a few LLMs to write a simple response to a question. Most of them were this way, writing paragraphs of blather even when I told it not to. The ones that didn't were all, without fail, formatted and worded like a Reddit post. Unsurprising because that's where the training data came from. All in all it cemented what I already knew that if you're smoothbrained enough to fall for AI hype then you're too dumb to operate an email client, or worse, have no shame and are willing to let everyone know you actually spend time on Reddit.
- Joined
- Aug 21, 2019
I am not anti-AI. I think AI is very useful. However, I write a lot of emails for work. There is nothing I have to write on a daily basis that takes more than a minute or two, and often times it feels like formulating a prompt and typing that for AI to write an email for me would take just as long as writing the thing myself. I've come to the conclusion that the AI tools built into things like email and text messaging programs are for people who cannot type faster than they speak. So while it may be useful if you are a person who uses speech to text to ask an AI to write an email, it's not terribly useful if you can type > 50 wpm with minimal spelling errors and put together a paragraph in your head on the fly. Basically, the dumbest people would find the greatest utility from this type of AI. I guess it can kind of work as an equalizer.
Last edited:
funwithshemales69
kiwifarms.net
- Joined
- Jan 6, 2025
The newgen poojeets are much much worse. Ive had to deal with some at services and uni stuff, probably would have to deal with some at my new job, theyre like older poojeets on steroids. Theyre so obsessed with ai that I wager theyd forget how to eat and shit in couple generations. They cant write a single word in english without AI and they make no attempt to be more skilled or learn something of value, everything is about credentialism to them and it has driven me insane sometimes, to share a world with these people who are non sentient. "Oh you know what a pointer is?" "Saar I dont know saar, I have 10 different certificates on Java and C saar!". Everything is credentialism, even their fucking english language skills is about credentialism, generate some shit out of gemini or chatgpt for an email or an essay and call it authentic work when they dont know what "persistent" is or how to spell it. Im starting to suspect that the entire country is a gigantic credentialist scam built on twigs and bubblegum. Sometimes I get anxious about being outclassed by these people cause credentialism does matter to some extent, the game is definitely rigged but Ill always take pride in the fact that I put full effort into my work and Im more educated and intelligent than these fuckers can ever aspire to be.
Tax: I hate this new google/google forks feature of auto-filtering search results by location. its a minor inconvenience but still annoying, knowing I have less control over my own browsing experience and I have more ways for people to spy on me.
CEO of Autism
kiwifarms.net
- Joined
- Oct 2, 2024
Catbox is now blocking me because... ???
I bet it has to do with me using a firefox based browser. I hate how more and more sites are steering you towards forced Chrome use. Soon you won't be allowed to browse any sites unless you're using goodgoybrowser™ on your real IP with adblock deactivated and previous ID check to prove you're a reel human bean.
I bet it has to do with me using a firefox based browser. I hate how more and more sites are steering you towards forced Chrome use. Soon you won't be allowed to browse any sites unless you're using goodgoybrowser™ on your real IP with adblock deactivated and previous ID check to prove you're a reel human bean.
- Joined
- Mar 4, 2019
I just tried with three of them and got in fine. VPN, fine. Maybe it's just you in particular?
- Joined
- Dec 7, 2020
I was doing some programming getting into search algorithms. I taught myself OpenCL which is like OpenGL, but for computational tasks that don't need a window or graphics context; this is all ran on the GPU. However you implement the algorithm is not important. What I found is that the best results with the different ranking algorithms I tired all put the best search results like at 2nd, 3rd, or 4th place in the rankings. This is because the algorithms are not very good and if there is one term in the query over represented in the searched documents or webpages it skews to the top of the results even-though the rest of the query is basically being ignored. I've experienced this for decades now using Yandex and Google and I never understood why the best results are always a few items down from from the top, where the top is the highest ranked result.
Creating a search index:
The documents put into the search index have to be processed and then indexed. The index only holds a table (how ever you represent it) that has the document (file path or reference number), the number of times the word appears in the document (popularity), and the word. The polarities are what are being ranked and the words are what are matched form the query.
1. Get text for all documents being indexed.
2. Delete all the words that are super common: all the prepositions and articles. This is done with a list of words to drop when counting words in the document.
3. Create a map (basically a table) of all the remaining words and their popularity for each document.
4. Store the document, word, and popularity in what becomes your Search Index. I did this with a Key Value DB and JSON, but you could do this different ways.
After the above steps all you have is a search index. To search it, you need to write a program that will compare words in the query with the popularities and return the documents ranked by popularity numbers. There are a lot of different ways to do this. There are a lot of different ranking algorithms, some more complex than others having special cases for words and special ways of processing queries. What I did was the simplest. Words in the query get put though the same processing and the documents and then words that remain are matched against the index, ranked by the popularity, and then the document (by reference ID) is returned. The interesting part is that I played around with different ranking algorithms I found online.
Processing the query:
1. Get query.
2. Delete all the words that are super common.
3. Create a matrix (table) to send to the GPU. This looks like each column is a word from the query, so we will have a value of 0 (zero) in some columns and others will have the popularity number of the word. The rows are the documents by reference ID. When you send this to the GPU the column and row labels are not sent with it, the order is maintained in this process and you keep the data used to create the table to turn it back into something comprehensible to the rest of the program. This step often is poorly represented online where I've seen it described. There is not any explanation given on how this matrix is created or how you know what the data in the matrix actually represents. Basically, it just the natural sorted order of the documents and their popularity values streamed into a vector, and then you know to divide the vector by the number of word popularities being compared to get the rows.
4. Send this vector to the GPU and tell the GPU to run the algorithm on it. This is done in OpenCL. Using OpenCL is not required and if this was all being done on CPU or single threaded the matrix step could be different and the algorithm implementation would look different.
5. After the GPU does maths, you get back a single vector, each number in the vector represents a row from the matrix and that row is then matched back with the document that was associated with the columns and then is sorted on the ranking number.
(There are some data normalisation steps and query preparation that I left out. Basically I made the query equal a document that has the sum of all popularities form the matched documents. I also wanted all the values to have a range between -1 to 1 when doing the calculations and then be represented from 0 to 1 when returned to the program. These steps are not essential and were really more of a way for me to get more comfortable with OpenCL.)
At the end you have documents denoted by ID numbers because it would be ridiculous to keep the string for the whole document as the label, obviously, and the rankings. The matrix is no longer kept around at this point and you just sort the documents by the values given from the GPU. You can now decided how you want to display this information.
I will post the source code for all the algorithms I tried. Sometimes the algorithms would represent things in descending order instead of ascending order. All the algorithms were changed so that the rank was inversely proportional to the match strength; best smallest number first worst biggest number last. Every algorithm had the same quagmire that they could be tricked by having too high a search term popularity in some documents and the algorithms didn’t try to find what document scored bast on all search term popularities provided. I found Manhattan Distance to give results that I liked best.
What I have done does not use any ML. In the future I may look into using ML for indexing and processing queries. AI chatbots are trained to be chatbots not matching and ranking algorithms. I don’t know what it would look like to train an AI for document ranking. Might be something to look into in the future.
The documents put into the search index have to be processed and then indexed. The index only holds a table (how ever you represent it) that has the document (file path or reference number), the number of times the word appears in the document (popularity), and the word. The polarities are what are being ranked and the words are what are matched form the query.
1. Get text for all documents being indexed.
2. Delete all the words that are super common: all the prepositions and articles. This is done with a list of words to drop when counting words in the document.
3. Create a map (basically a table) of all the remaining words and their popularity for each document.
4. Store the document, word, and popularity in what becomes your Search Index. I did this with a Key Value DB and JSON, but you could do this different ways.
After the above steps all you have is a search index. To search it, you need to write a program that will compare words in the query with the popularities and return the documents ranked by popularity numbers. There are a lot of different ways to do this. There are a lot of different ranking algorithms, some more complex than others having special cases for words and special ways of processing queries. What I did was the simplest. Words in the query get put though the same processing and the documents and then words that remain are matched against the index, ranked by the popularity, and then the document (by reference ID) is returned. The interesting part is that I played around with different ranking algorithms I found online.
Processing the query:
1. Get query.
2. Delete all the words that are super common.
3. Create a matrix (table) to send to the GPU. This looks like each column is a word from the query, so we will have a value of 0 (zero) in some columns and others will have the popularity number of the word. The rows are the documents by reference ID. When you send this to the GPU the column and row labels are not sent with it, the order is maintained in this process and you keep the data used to create the table to turn it back into something comprehensible to the rest of the program. This step often is poorly represented online where I've seen it described. There is not any explanation given on how this matrix is created or how you know what the data in the matrix actually represents. Basically, it just the natural sorted order of the documents and their popularity values streamed into a vector, and then you know to divide the vector by the number of word popularities being compared to get the rows.
4. Send this vector to the GPU and tell the GPU to run the algorithm on it. This is done in OpenCL. Using OpenCL is not required and if this was all being done on CPU or single threaded the matrix step could be different and the algorithm implementation would look different.
5. After the GPU does maths, you get back a single vector, each number in the vector represents a row from the matrix and that row is then matched back with the document that was associated with the columns and then is sorted on the ranking number.
(There are some data normalisation steps and query preparation that I left out. Basically I made the query equal a document that has the sum of all popularities form the matched documents. I also wanted all the values to have a range between -1 to 1 when doing the calculations and then be represented from 0 to 1 when returned to the program. These steps are not essential and were really more of a way for me to get more comfortable with OpenCL.)
At the end you have documents denoted by ID numbers because it would be ridiculous to keep the string for the whole document as the label, obviously, and the rankings. The matrix is no longer kept around at this point and you just sort the documents by the values given from the GPU. You can now decided how you want to display this information.
I will post the source code for all the algorithms I tried. Sometimes the algorithms would represent things in descending order instead of ascending order. All the algorithms were changed so that the rank was inversely proportional to the match strength; best smallest number first worst biggest number last. Every algorithm had the same quagmire that they could be tricked by having too high a search term popularity in some documents and the algorithms didn’t try to find what document scored bast on all search term popularities provided. I found Manhattan Distance to give results that I liked best.
What I have done does not use any ML. In the future I may look into using ML for indexing and processing queries. AI chatbots are trained to be chatbots not matching and ranking algorithms. I don’t know what it would look like to train an AI for document ranking. Might be something to look into in the future.
C:
// Open CL Kernel
/* // Cosine Similarity */
/* size_t i = get_global_id(0); */
/* float */
/* dp = 0.0f, */
/* n1 = 0.0f, */
/* n2 = 0.0f; */
/* for (size_t ii = 0; ii < elements; ii++) { */
/* float */
/* v1 = matrix[i * elements + ii], */
/* v2 = query[ii]; */
/* n1 += v1 * v1; */
/* n2 += v2 * v2; */
/* dp += v1 * v2; */
/* } */
/* scores[i] = 1 - (dp / (sqrt(n1) * sqrt(n2))); */
// Manhattan Distance
size_t i = get_global_id(0);
scores[i] = 0.0f;
for (size_t ii = 0; ii < elements; ii++) {
float
v1 = matrix[i * elements + ii],
v2 = query[ii];
scores[i] += fabs(v1 - v2);
}
/* size_t i = get_global_id(0); */
/* scores[i] = 0.0f; */
/* for (size_t ii = 0; ii < elements; ii++) { */
/* float */
/* v1 = matrix[i * elements + ii], */
/* v2 = query[ii]; */
/* scores[i] += 1 / fabs(v1 - v2); */
/* } */
/*
// Vector Inner Product
size_t i = get_global_id(0);
scores[i] = 0.0f;
for (size_t ii = 0; ii < elements; ii++) {
float
v1 = matrix[i * elements + ii],
v2 = query[ii];
scores[i] += v1 * v2;
}
// scores[i] = 1 / scores[i];
*/
/*
// Tanimoto Coefficient
size_t i = get_global_id(0);
float
dp = 0.0f,
n1 = 0.0f,
n2 = 0.0f;
for (size_t ii = 0; ii < elements; ii++) {
float
v1 = matrix[i * elements + ii],
v2 = query[ii];
n1 += v1 * v1;
n2 += v2 * v2;
dp += v1 * v2;
}
scores[i] = 1.0f - (dp / (sqrt(n1) + sqrt(n2) - dp));
*/
}
)";
Mad as a Swan
kiwifarms.net
- Joined
- Aug 16, 2025
Welcome to the pajeetistan/HR world where a shitty paper is more valuable than what you know.
I have the same problem applying in tech: broad portfolio of projects and demonstrable skills? Nobody cares. Pajeet with 20 certs? Pass! It's clear that recruiters don't even care about quality of the candidates; they just filter through an algo and the algo goes: "Pajeet, BIPOC check! 10 certificats? Skills check! Github PR every day? Active coder pass!"
You WILL be outclassed by these people and they WILL manage you and they WILL be smug about it and wave their credentials around like they're big deal, and higherups (also pajeets) will look at these credentials, and back at you, and side with them because they have more paper.
Wanted to rant about this. New phone. Try to log in into something; give username, password, 2FA in some cases... Denied anyway. Why? Always "new device" or "unauthorized" or whatever. And it's for some retarded social media like X or some shit that you use casually.
Whatever happened to just logging in using your username and password? Normies have ruined it to the point you can't even trust them with a password; you need to secure a device to secure a password manager to log in.
- Joined
- Nov 21, 2020
Do you want to use a Passkey? It's more secure.
No.
Are you sure, it's much better, it's tied to your device.
Fuck no.
Hey, how about setting up a passkey for your account.
Fuck off.
Hey, we automatically tried to setup a passkey for you but it failed.
Yes, because I turned that shit off in the settings. Fuckers.
- Joined
- Sep 1, 2019
The robot vacuum my parents gave me tried forcing me to make an account and pass key to use the wifi features for the vacuum. They think I will give them all my information instead of just manually pressing a button.