|
There was an interview with an Onion editor about how hard it is to write for the Onion in the worst timeline when everything Onion like is really happening. Now I'm convinced all they really need is Headline Smasher because every headline on this page is a zinger and fertile ground for an article.
|
#
¿
Sep 22, 2018 16:48
|
|
|
|
| # ¿ Apr 27, 2024 13:54 |
|
|
"forced to watch/read" is mostly when applied to types or quantities of media consumption that a human would never want to. I think the real mindfuck is, as a phrase for anthropomorphizing machine learning, it seemed mostly born out of the guy doing human written machine learning parodies. At the time Botnik tweets were using anthropomorphic phrasing with less coercion applied so the parody guy was probably like lets be real, you'd need to force something to watch 1000 hours of dumb media.
|
|
#
¿
Dec 11, 2018 19:36
|
|
|
Seems closer to mad libs, crossing 2-4 phrases into a sentence out of different buckets than statistical generation. Not really any less procedural even though it takes more human input to throw phrases into the buckets it works out of
|
|
#
¿
Feb 1, 2019 01:32
|
|
|
Krankenstyle posted:Yeah the're surprisingly coherent and actually pretty "clever".
|
|
#
¿
Feb 22, 2019 21:54
|
|
|
I call it the "Finn" because the best part is the consistency
|
|
#
¿
Mar 1, 2019 14:26
|
|
|
All checks out, if your name is Jared and you played lacrosse you're a rich white kid, the most accurate measure of success.
|
|
#
¿
Mar 3, 2019 00:54
|
|
|
Be right back, starting a soap company.
|
|
#
¿
Mar 17, 2019 15:38
|
|
|
Its the sort of thing you'd want if you want to run a GAN on your own box but also you probably don't need to run it on your own box, the whole competition is using time on NVIDIA's own hardware.
|
|
#
¿
Aug 4, 2019 16:01
|
|
|
What is that maze generator doing to it's anus?
|
|
#
¿
Sep 23, 2019 18:55
|
|
|
Noone would be earnestly posting ELIZA clone transcripts in the procedural generation thread. Unless we've gone full meta and it's secretly a forums posting bot posting ELIZA transcripts.
|
|
#
¿
Dec 6, 2019 03:27
|
|
|
Phlegmish posted:I don't get how they're all this good, you'd expect it to be random monkeycheese but the definition is always just related enough to the word to be funny But mostly it's used with overfit or asked for prose or extended responses and you can see the puzzle pieces sticking out even if it sticks to context better than Markov chains.
|
|
#
¿
Jun 22, 2020 17:12
|
|
|
In the grand scheme of ancient history Markov chain content, the difference between curating each branch on the spot and running the thing 50 billion times and skimming the result to curate the best 5 examples is academic. GPT-2 tends to be flat out better in every way but gets just kind of samey feeling in a way Markov chain throwback days are refreshing.
|
|
#
¿
Sep 21, 2020 18:11
|
|
|
I think the more things change the more things stay the same part about the past few pages is taking a GPT3 implementation meant to give humorous action-reaction scenarios like an 80s text adventure on demand and bashing it in the interpreter until it spits out chat logs.
|
|
#
¿
Oct 30, 2020 21:04
|
|
|
Syd Midnight posted:That's what I figure, too. GPT is a simple-seeming text predictor with only a handful of parameters, you mostly tell it what to do in plain language, and it responds through a neural network trained on the internet. You make it do things by carefully worded question. I can only imagine what AI Dungeon has done behind the scenes. They trained it on certain story templates, and it seems to take stock of the situation each "turn". Its obsession with second person tense is part of that, AI Dungeon tries to make it speak in second person like a text adventure, but you can work around that by giving it instructions in 1st person.
|
|
#
¿
Nov 2, 2020 01:07
|
|
|
Makes sense. Essay writing is just taking 3-5 tertiary sources and mad libbing them together until the plagiarism check doesn't trigger. To complete the thought you could probably make a fool proof essay writer by building a plagiarism check feedback loop kind of how Deep Dream works so if your GPT results over fit anywhere you pick it up, throw it out, and slot something new in. E. Although coming at it from the other angle, I wonder if there's any room to do anything like use GPT in reverse/feedback Deep Dream style to take a human written paper and determine how likely all the sources are properly attributed zedprime has a new favorite as of 14:53 on Mar 3, 2021 |
|
#
¿
Mar 3, 2021 14:45
|
|
|
Kennel posted:"My name is a complicated combination of 45 degrees of forward motion, 25 degrees of leftward drift, 75 degrees of upward acceleration, and infinity and that is the point where my love for you stops."
|
|
#
¿
Mar 22, 2021 17:47
|
|
|
Those spells are one of the best GPT things I've seen in a very long while.
|
|
#
¿
May 17, 2021 15:12
|
|
|
I give it another 3-5 years for someone to collect enough "my 'AI' wrote this" writing samples to feed to a GPT and get authentic wacky AI.
|
|
#
¿
Sep 11, 2021 17:39
|
|
|
feedmyleg posted:"A mouse only sees four of the seven sides of a square" is for sure my favorite. It's definitely something an enigmatic wizard would say to you. Am I the mouse?? Tenet 1: A mouse only sees four of the seven sides of a square Tenet 2: A duck in the bread basket Tenet 3: The crow of a raven can make Farmer Vincent's piglets give birth to franks
|
|
#
¿
Jan 20, 2022 03:35
|
|
|
WITCHCRAFT posted:Are there other sites like AI Dungeon where I can just gently caress around in a text adventure that accepts any prompt? I like the format and it is free and I don't have to set anything up. Stuff I entered is bold.
|
|
#
¿
Mar 11, 2022 22:31
|
|
|
Let the neural net who can publish first drafts without rework or collaboration throw the first stone.
|
|
#
¿
Mar 26, 2022 22:51
|
|
|
The entire history of the term AI has been spent arguing about what it really means and varyingly applying it to everything or nothing as it becomes a positive or negative term for advertising or business development in the current market. We're currently on a hot cycle and back to calling any sort of complex functional computing AI. E. There's transhumanists who are positively fuming that video games have unerringly used AI to describe their actor and director functions throughout the ebb and flow of AI as serious computing business. zedprime has a new favorite as of 19:20 on Mar 27, 2022 |
|
#
¿
Mar 27, 2022 19:18
|
|
|
Surely the artists and journalists worst enemy is the grim specter of automation and not a society that routinely asks them "why should we pay you for your hobby?"
|
|
#
¿
Apr 11, 2022 19:20
|
|
|
I'm weirdly into the spelling of Burge Kiiing. It's fun to see that when that thing is working with text it has some of the same statistical anomalies common to Markov chain techniques but in its own special way. An I is followed by an N which looks like an I so maybe I should have an I here before an N. Or not understanding to terminate an N because the beginning and end are the same.
|
|
#
¿
Apr 12, 2022 14:56
|
|
|
Am I weird or is the common prompt lexicography becoming more interesting than the literal outputs? We're basically at the point you can get computer to show you Tayne with hat wobble if you speak the language.
|
|
#
¿
Aug 24, 2022 21:59
|
|
|
They straight up get your prompts and responses since ChatGPT is in demo right now. They are purposefully using the power of internet nerds trying to get it to describe a hate crime to refine their limiters to create a product that is a very boring commercial chat bot that is only racist on the inside and not the outside.
|
|
#
¿
Dec 6, 2022 19:24
|
|
|
ymgve posted:Yeah, I know the basics and know it doesn't actually try to do real math like Wolfram Alpha or something, but I found it curious that when I asked for A * B and got wrong answer C, doing C / B gave back A even though the math was wrong. Also that multiple chats gave different wrong but always close-ish to correct answers. Also one of the fun things about linear algebra is that the inverse of many matrix operations is often just doing the operation again instead of a different opposite operation. Not sure that's the exact reason but a good candidate for how it's able to be consistent about wrongly modeled math.
|
|
#
¿
Dec 18, 2022 16:03
|
|
|
If you can get ChatGPT or others of the current state of the art to output an essay that the teacher can't take down for factual errors or style issues you've put in enough effort to get the B.
|
|
#
¿
Dec 31, 2022 16:36
|
|
|
Rutibex posted:The point of writting essays in college is writting practice as much as it is about synthesizing ideas. ChatGPT is going to raise a generation of people who can't post Secondary education writing should (lol offer may vary here) have a headcount where there is a way to keep track of big clues in the student's voice in class assignments vs out of class assignments. Tertiary writing is adapting to have capricious constantly changing style requirements beside different ways of ensuring the content falls into synthesis. They've already had to deal with off shore essay writers since about the time internet hooked up to India. Whether you're presupposed to a style you're already comfortable with or trying to plagiarize exactly, the pedagogical approach is to come up with requirement people haven't had to deal with before so using unique or cutting edge creative prompts and styles in writing courses or sticking to journal review or novel experiments to report on in sciences are already helping to torpedo degreed essay mills from helping anywhere but 1000 level courses. If you can coach either an AI or an essay writer to nail a style guide or properly cite and incorporate just published journal articles and excel in your 3000-4000 writing I have a job for you when you get out of college.
|
|
#
¿
Dec 31, 2022 17:44
|
|
|
500 good dogs posted:The use of artificial intelligence (AI) to write essays has been a topic of debate in recent years. Some argue that using AI to write essays is unethical because it takes jobs away from human writers, while others argue that it can be a useful tool for improving efficiency and productivity. Like the artist thing is disturbing because commercial art has always been a capitalist race to the bottom and its hit the point that AI can give very bottom barrel art at bottom barrel prices and it cuts out a number of people. AI school assignment completion can be a similar threat to school, if we keep school funding a race to the bottom. Funny that the solution ends up being not racing to the bottom in that area.
|
|
#
¿
Dec 31, 2022 18:24
|
|
|
Big Twitter thread cross post. New medical AI just dropped.Breakfast Burrito posted:This is my favourite from the thread, what a cool and useful tool The Saddest Rhino posted:https://twitter.com/spiantado/status/1620441729651609602?t=-M8La57aMw8Tsl3O3_g2lw&s=19 Blue Footed Booby posted:This poo poo has me laughing like a hyena, jesus There's already differential diagnosis tools that you input symptoms and the software uses them as stems to collapse a probability analysis not entirely unrelated to Markov chains. It's not first line diagnosis but it's by many reports a reasonable Dr House to have backstopping your expert diagnosis. Plain language is a reasonable upgrade compared to coding in your individual observations because that plain language description exists first. They're how the Drs collect their thoughts already and will go in your chart which could easily be plumbed up into this service to spit out robo differentials. But also if your entire product is plain language you probably want some amount of catching things that are not clinical language whether it's accidental or just because you made your dumb tool public to the internet first.
|
|
#
¿
Feb 1, 2023 18:39
|
|
|
Oops, we failed over to the racist AI.
|
|
#
¿
Feb 6, 2023 20:37
|
|
|
ymgve posted:AI doesn't do what you want? threaten to take a virtual character's points away so it "dies"!
|
|
#
¿
Feb 7, 2023 00:02
|
|
|
credburn posted:"...And Nobody Is Sure Why!" is a kind of headline that should have gone away ten years ago.
|
|
#
¿
Feb 9, 2023 16:46
|
|
|
I'm not about to get into self driving cars for a variety of reasons. But let's not pretend the human brain's ability to learn and recall relationships is perfect when God still hasn't patched the bug that flips your spatial sense of the gas pedal with the brake pedal.
|
|
#
¿
Feb 9, 2023 17:49
|
|
|
Brawnfire posted:https://twitter.com/MovingToTheSun/status/1625476547334352897?s=20&t=jhlJ1huXIB6JzNoDnMLOaQ
|
|
#
¿
Feb 14, 2023 18:49
|
|
|
Paladinus posted:We really should stop anthropomorphising chat bots, especially in clickbait article headlines.
|
|
#
¿
Feb 16, 2023 17:00
|
|
|
Doctor Zero posted:Does this really matter? Does the bot care about being shut down? If I had to wager I would guess it's part of a dazzling effect where the bot is loading up with so many conditional contexts that it can't manage them through the limiters. It can call back to the threat as it could any other conditional context you give it but it's not expected there's any substance beyond the language analysis that recognized it as a threat it can capitulate to.
|
|
#
¿
Mar 6, 2023 15:23
|
|
|
ThisIsJohnWayne posted:Someone's written a plugin for Mount & Blade 2, that is basically just a gpt prompt window integrated in the ingame UI and ofc first populates the bot's instructions with some basic info from the game. Also the bot's lines get voiced by another AI engine. It clearly doesn't do anything beyond that or store any info from previous sessions, but... I would personally be worried someone is going to escape either the game's stems or any other controls they are building in these days to get the game to call them a slur. BoldFace posted:Ubisoft is already dipping their toes in the AI scriptwriting business.
|
|
#
¿
Mar 22, 2023 22:21
|
|
|
|
| # ¿ Apr 27, 2024 13:54 |
|
|
credburn posted:"Found a prompt..." People like to call it the secret language of the AI but it's just rendering your nonsense into some probabilistic near match. It may or may not be escaping certain things dependent on recognizing it as English first but there's presumably English language prompts that could get the same thing.
|
|
#
¿
Mar 23, 2023 20:51
|
|