|
why is everyone so wound up over artificial insemination all of a sudden
|
# ? Mar 30, 2023 17:09 |
|
|
# ? Apr 28, 2024 15:20 |
|
Shame Boy posted:he's an "AI researcher" in that he writes lots and lots of words about how scary it would be if we made the robot devil, and how you should give him and his organization money so we can make robot god before that happens to fight the robot devil dont forget about using the fanfiction and nutso writings to gently caress 6 women at once like a saudi prince and grift billionaires
|
# ? Mar 30, 2023 17:11 |
|
bob dobbs is dead posted:dont forget about using the fanfiction and nutso writings to gently caress 6 women at once like a saudi prince and grift billionaires thought for a moment this was an admiring cremnob driveby
|
# ? Mar 30, 2023 17:12 |
|
more like chudkowski imo
|
# ? Mar 30, 2023 17:18 |
|
Progressive JPEG posted:why is everyone so wound up over artificial insemination all of a sudden ai is how musk has seized various means of production.
|
# ? Mar 30, 2023 17:19 |
|
Mr. Nice! posted:ai is how musk has seized various means of production.
|
# ? Mar 30, 2023 17:21 |
|
Roko's basilisk was funny, but it was also directly responsible for $TWTR printing at $54.20/share and that's way funnier
|
# ? Mar 30, 2023 17:34 |
|
mystes posted:usually when people talk about "seizing" the means of production they mean "take hold of" not "become stuck or jammed" i was making a joke about his myriad progeny and preferred form of reproduction.
|
# ? Mar 30, 2023 17:37 |
|
Beeftweeter posted:more like chudkowski imo i was washing dishes 10 minutes ago and this thought intruded over everything else and I’m so mad I got beat
|
# ? Mar 30, 2023 18:05 |
|
Fart Sandwiches posted:i was washing dishes 10 minutes ago and this thought intruded over everything else and I’m so mad I got beat i kinda hate that i find "thought of a great post while doing something else and held onto it real tight until i could rush back to my computer to post it, only to find someone beat me to it" relatable
|
# ? Mar 30, 2023 18:06 |
|
Fart Sandwiches posted:i was washing dishes 10 minutes ago and this thought intruded over everything else and I’m so mad I got beat
|
# ? Mar 30, 2023 18:16 |
|
thinkin about those posts
|
# ? Mar 30, 2023 18:17 |
|
happens to literally everyone, normal people call it shower thoughts
|
# ? Mar 30, 2023 18:18 |
|
what are the useful applications of LLMs, everything I've seen so far is either frivolous (AI dungeon, smut) or actively harmful (blogspam, misinformation, writing buggy code)
|
# ? Mar 30, 2023 18:18 |
|
active harm
|
# ? Mar 30, 2023 18:20 |
|
BMan posted:what are the useful applications of LLMs, everything I've seen so far is either frivolous (AI dungeon, smut) or actively harmful (blogspam, misinformation, writing buggy code) everything else is automated fascism
|
# ? Mar 30, 2023 18:27 |
|
qirex posted:businesses are interested because the dream is they can outsource anything in the first two tiers of customer contact/service to a computer and maybe qa too so actively harmful, then
|
# ? Mar 30, 2023 18:29 |
|
currently you cant get them to say no to poo poo so the counterplay is just ask for 99% discounts or some poo poo
|
# ? Mar 30, 2023 18:30 |
|
Shame Boy posted:so actively harmful, then not to investors who will reap a quick reward for eliminating all those call centers in the philippines!
|
# ? Mar 30, 2023 18:32 |
|
bob dobbs is dead posted:currently you cant get them to say no to poo poo so the counterplay is just ask for 99% discounts or some poo poo yeah there are not going to be any practical applications until they prove they have solved both the hallucination problem and their tendency to yield to user insistence even on questions of fact or policy like do you really want people using DAN-style tricks on your customer support
|
# ? Mar 30, 2023 18:40 |
|
BMan posted:what are the useful applications of LLMs, everything I've seen so far is either frivolous (AI dungeon, smut) or actively harmful (blogspam, misinformation, writing buggy code) in immediate things, better machine translation is a great thing, and that is very much flowing out of this stuff now. and there's some other light things like that to add to the pile. beyond that it is to a great extent still unknown. i personally kind of refuse to believe that we wont be able to bend computers being able to usefully process natural language into some very great things in the future. making many things ("systems" very broadly interpreted) literally self-explanatory in a very human-oriented way. as a very specific example i've used locally (so a bunch of specific systems in play which i don't know how they map onto anywhere else) seeking healthcare during an immigration process, which involves a billion forms, bits of information, checklists, and different systems that interact primarily *through* you. notably there is no human worker at precisely your intersection of things. but there is some hope that an llm could be puppeteered into putting all these disparate aspects into e.g. one coherent letter, in your preferred language, and be able to update instructions or guide towards clarifications (i.e. looping in the right people in each subsystem) when these things don't work out. all very soft, but the point kind of is that the promise is in very soft applications, not really (the way i see it at least) in any big killer app that'll make a trillion dollars. also i should not that "puppeteer" here hides a real problem, because just throwing everything into a prompt and hoping for the best is not the way i think anything of the sort can be done.
|
# ? Mar 30, 2023 18:43 |
|
bob dobbs is dead posted:currently you cant get them to say no to poo poo so the counterplay is just ask for 99% discounts or some poo poo oh, sydney will absolutely tell you to gently caress off
|
# ? Mar 30, 2023 18:46 |
|
Cybernetic Vermin posted:in immediate things, better machine translation is a great thing, and that is very much flowing out of this stuff now. and there's some other light things like that to add to the pile. the thing about your example is it could also just as easily be used to deny people services that they do qualify for by being deliberately obtuse. I'm not concerned about chatgpt becoming self-aware, I worry that it will be used to increase the cruelty of our existing systems and distance the decision makers who control it from the general public. from my point of view it seems a lot closer to being able to do that than to actually help people
|
# ? Mar 30, 2023 18:47 |
|
qirex posted:the thing about your example is it could also just as easily be used to deny people services that they do qualify for by being deliberately obtuse. I'm not concerned about chatgpt becoming self-aware, I worry that it will be used to increase the cruelty of our existing systems and distance the decision makers who control it from the general public. from my point of view it seems a lot closer to being able to do that than to actually help people oh, for sure. ultimately i am paid to be pretty adjacent to this stuff, and for the sake of my sanity i have had to accept i can't really affect the ultimate outcomes. i will call this at least a *possible* positive outcome though. as such i also don't envision a single *decision* being made in the llm, it is not really within the models capabilities. that's also part of why i talk about some unknown procedure for "puppeteering" the models, because i have fairly low faith in prompting a fine-tuned model into any complex task (which this certainly is). there needs to be some actual logic in the loop to ensure the intended thing does get expressed. and how to add such logic is not yet well understood.
|
# ? Mar 30, 2023 18:54 |
|
you use bayes nets, its been understood for decades. thats how the waymo poo poo works
|
# ? Mar 30, 2023 18:56 |
|
Mr. Nice! posted:i was making a joke about his myriad progeny and preferred form of reproduction.
|
# ? Mar 30, 2023 19:03 |
|
rotor posted:i dont really read this thread much but ok well as long as we can all agree the data centers should burn, im cool when did they make you a mod?
|
# ? Mar 30, 2023 20:13 |
|
BMan posted:what are the useful applications of LLMs, everything I've seen so far is either frivolous (AI dungeon, smut) or actively harmful (blogspam, misinformation, writing buggy code) massive intellectual property theft on an unprecedented scale the ultimate goal is to be able to feed any copyrighted work into their LLM without permission the moment it is released and then be able to say “write me a novel in the style of Steven King” and own the copyright on the LLM and anything it produces
|
# ? Mar 30, 2023 20:19 |
|
its a side effect of the bitter lesson. the algorithms are commodity, the hardware is even more commodity, the data is still basically commodity if you do the ip theft. openai is just a less ripe target than the tech majors so they went and did that ip theft
|
# ? Mar 30, 2023 20:22 |
|
pseudorandom name posted:massive intellectual property theft on an unprecedented scale it devalues everything it touches, it's just not very good at it. normally that might mean it wouldn't be used because it produces inferior results, but practically speaking it means that everything will get just that much shittier as this becomes the new baseline
|
# ? Mar 30, 2023 20:24 |
|
i mean, first fusillade of lawsuits have also been fired. see what the courts say
|
# ? Mar 30, 2023 20:26 |
|
pseudorandom name posted:massive intellectual property theft on an unprecedented scale the criticism is absolutely true, but the claimed "goal" is either way way overestimating what the models can do or underestimating what the courts can understand on that theme, this post has been kicking around the back of my mind for a bit: Shame Boy posted:a much larger portion of the furry community has gotten real fuckin' mad at the AI porn people cuz if nothing else furries respect the hell out of their artists so that's nice i guess yep. that is true. and kind of to the point, decent people should get a chance to understand what is happening, and hopefully attitudes will be decent as a result. e: and, yes, here the decent people are the furry porn enthusiasts
|
# ? Mar 30, 2023 20:27 |
|
jammyozzy posted:This is gonna backfire when the only person lifeless and creepy enough for the mii creator to make a reasonable facsimile of is Zuck This is from way back but is there a term for “over dogfooding?” Overfitting? I guess it’s just bias. Don’t let the alien people like Zuck dog food the product or it won’t be fit for normal people. Also, excited for how LLM is going to turn the world into a Borges story. Basically library of Babel but backwards: only the books that have conceivable meaning previously existed, but now AI is going to be filling the shelves with gibberish. Edit: Imagining cultural heat death. Anything made by a human vastly separated by AI emptiness. Pulcinella fucked around with this message at 21:26 on Mar 30, 2023 |
# ? Mar 30, 2023 21:22 |
|
pseudorandom name posted:massive intellectual property theft on an unprecedented scale right now you can pay someone to write a novel in the style of Steven King and it'll always be a million times better than anything some neural net would spit out. there's no ai pattern matching magic that'll be able to give you something better than what a human could produce with the same prompt. will it let you create vast quantities of them very quickly? sure, but the reason lovely knock offs don't sell well isn't because there are too few of them.
|
# ? Mar 30, 2023 21:33 |
|
as a preface, i completely agree that the vast majority of llm applications are scams, actively harmful, or both i've been in a few discussions where i see a glimmer of a valuable application category: solving the "blank page" problem for creative and/or intellectual work. for example, you could imagine a data analysis tool that uses llms to churn over metadata and come up with suggestions for queries you might want to run. some of them will be bullshit, some might be helpful ideas. keep the human in the driver's seat, shaping and refining, while the machine spitballs and free-associates responsible applications of the technology basically need oversight and copy-editing of anything the model pieces together, so lean into that angle instead of trying to paper over it or remove the human from the loop entirely
|
# ? Mar 30, 2023 21:35 |
|
Pulcinella posted:This is from way back but is there a term for “over dogfooding?” Overfitting? I guess it’s just bias. Don’t let the alien people like Zuck dog food the product or it won’t be fit for normal people. what you're describing here is more the death of the *internet* though. it may no longer work or make sense to just find any text that says something and figure that it is knowledge or content. but that may be just as well, social context already mattered, this imagined future just fully poisons the realm of finding and accepting non-contextual random nonsense.
|
# ? Mar 30, 2023 21:38 |
|
trusting poo poo cuz its written didnt work when gilgamesh roamed ur, why would you say it stopped working now
bob dobbs is dead fucked around with this message at 21:55 on Mar 30, 2023 |
# ? Mar 30, 2023 21:39 |
|
Cybernetic Vermin posted:the criticism is absolutely true, but the claimed "goal" is either way way overestimating what the models can do or underestimating what the courts can understand Chalks posted:right now you can pay someone to write a novel in the style of Steven King and it'll always be a million times better than anything some neural net would spit out. they don’t care, they’re venture capitalist scum and they’re gambling that it’ll work and they’ll get away with it and the profits will be astronomical
|
# ? Mar 30, 2023 21:40 |
|
ai has continually had absolute dog poo poo margins for pure software play. like 40, 50%
|
# ? Mar 30, 2023 21:41 |
|
|
# ? Apr 28, 2024 15:20 |
|
saas routinely gets 90% gross margins, for contrast 1, 10 million a pop for training adds up quite fast
|
# ? Mar 30, 2023 21:42 |