|
the real danger of llms is a further deepening of the already-present trust vacuum that exists on the internet, only now instead of going "well i don't know if the person who made this was telling the truth or not" it becomes "well i don't know if this was made by a person or not". i doubt this is really that big a deal in the long run since the truthiness of a statement is of much greater importance than who made it
|
#
?
May 30, 2023 17:57
|
|
|
|
| # ? May 12, 2024 14:29 |
|
|
quiggy posted:the real danger of llms is a further deepening of the already-present trust vacuum that exists on the internet, only now instead of going "well i don't know if the person who made this was telling the truth or not" it becomes "well i don't know if this was made by a person or not". i doubt this is really that big a deal in the long run since the truthiness of a statement is of much greater importance than who made it this is obviously a problem but i think it's frankly only slightly worse than what the internet has been for the last 8 years
|
|
#
?
May 30, 2023 17:59
|
|
|
gpt rollout for spam was complete by end of 2018
|
|
#
?
May 30, 2023 17:59
|
|
|
Not a Children posted:lmao the specificity of a 6 month pause is enough to radiate "we're full of poo poo" that one was specifically proposed so that elon's new company wouldn't be so far behind. it's incredibly obvious most of these people are 100% not genuinely concerned about this. like CV was saying, they know better. it's in openai and deepmind's interests to play this up though — they want people to think their products are more than "just" LLMs because it's driving an absolute frenzy of investment. they probably figure the odds of any real regulation are pretty slim, and even then wouldn't impact their actual products anyway, since they're chatbots it's telling that openai threatened to leave the european union entirely when they proposed extremely milquetoast regulations on LLMs
|
|
#
?
May 30, 2023 18:00
|
|
|
Intrusive thought of how fuckin dire college students' bibliographies/citations will be this yearBeeftweeter posted:that one was specifically proposed so that elon's new company wouldn't be so far behind. it's incredibly obvious Aren't they already speedrunning the "coordinate with congress to create an obscene barrier to entry for all competitors" playbook
|
|
#
?
May 30, 2023 18:00
|
|
|
mediaphage posted:this is obviously a problem but i think it's frankly only slightly worse than what the internet has been for the last 8 years yeah exactly, i don't think the problem of grampa reading on facebook that the gays are putting hormones in the 5g vaccine or w/e is gonna change significantly when the author of that post becomes an ai instead of a person
|
|
#
?
May 30, 2023 18:00
|
|
|
just pull out the blue books again
|
|
#
?
May 30, 2023 18:00
|
|
|
bob dobbs is dead posted:gpt rollout for spam was complete by end of 2018 And current gpt is much more convincing, yes
|
|
#
?
May 30, 2023 18:14
|
|
|
bob dobbs is dead posted:just pull out the blue books again
|
|
#
?
May 30, 2023 18:16
|
|
|
Improbable Lobster posted:And current gpt is much more convincing, yes great now when they threaten to send videos of me jacking off to all my friends and family, the grammar will be better
|
|
#
?
May 30, 2023 18:21
|
|
|
i hope that the ai hype is the last gasp of the tech bubble trying to demonstrate something actually innovative and justify the investments and valuations of the companies involved. so maybe we get lucky and there will be a slurp juice moment for the ai hype and reality asserts itself just a tiny little bit
|
|
#
?
May 30, 2023 18:21
|
|
|
The minute lawyers tried to use it they submitted made up case law, seems pretty bad
|
|
#
?
May 30, 2023 18:22
|
|
|
mediaphage posted:great now when they threaten to send videos of me jacking off to all my friends and family, the grammar will be better fortunately it's not too long before you could plausibly claim that the blackmail jerkoff video is also an AI fake
|
|
#
?
May 30, 2023 18:25
|
|
|
mediaphage posted:great now when they threaten to send videos of me jacking off to all my friends and family, the grammar will be better jokes on them i have no friends or family
|
|
#
?
May 30, 2023 18:26
|
|
|
i think deepfake videos are more of a problem than text tbh. they're getting pretty convincing
|
|
#
?
May 30, 2023 18:27
|
|
|
lol
|
|
#
?
May 30, 2023 18:28
|
|
|
yep. 110% marketing hype bullshit
|
|
#
?
May 30, 2023 18:30
|
|
|
[quote from guy who invented the Torment Nexus] we should really have better regulations on Torment Nexuses
|
|
#
?
May 30, 2023 18:31
|
|
|
haveblue posted:fortunately it's not too long before you could plausibly claim that the blackmail jerkoff video is also an AI fake yeah i was thinking about this a little while ago, as deepfake porn becomes more common and easy to make and convincing, it kinda provides this layer of plausible deniability that might take a lot of the teeth out of revenge porn
|
|
#
?
May 30, 2023 18:31
|
|
|
weird, the text of your post is not showing up when i hit quote anyway, EU: ok, we will put some restrictions on the torment nexus sam altman: what the gently caress. i meant other people's torment nexus
|
|
#
?
May 30, 2023 18:39
|
|
|
the forums probably think that's an actual quote that shouldn't be double-quoted jeffery plz
|
|
#
?
May 30, 2023 18:41
|
|
|
i think llms and generative ai will significantly change the way we interact with computers in the next 5 years the same way the iphone put a smartphone in everyone's hands why would my grandma need to click some buttons, type on a keyboard or learn some basic virtual postcard template program when she could just talk to her computer instead? laughing at the idea of the ceo of openai begging governments to regulate them though, that's just fomo investor hype bullshit go play outside Skyler fucked around with this message at 18:46 on May 30, 2023 |
|
#
?
May 30, 2023 18:43
|
|
|
haveblue posted:the forums probably think that's an actual quote that shouldn't be double-quoted
|
|
#
?
May 30, 2023 18:48
|
|
|
go play outside Skyler posted:i think llms and generative ai will significantly change the way we interact with computers in the next 5 years the same way the iphone put a smartphone in everyone's hands yeah i agree, though i'm not sure about the timeline. openai has also done some stuff as regards voice recognition and synthesis iirc. i don't think that the result is necessarily going to be some dire lack of trust that currently exists, though, just make the current internet potentially more accessible
|
|
#
?
May 30, 2023 18:48
|
|
|
i do see this stuff as a huge interface improvement but the thing is it still just wholely makes up poo poo so the interface is useless
|
|
#
?
May 30, 2023 18:48
|
|
|
yeah that is still the problem. we need confidence the AI isn’t just making poo poo up for it to have any chance of changing the world for the better
|
|
#
?
May 30, 2023 18:52
|
|
|
Share Bear posted:i do see this stuff as a huge interface improvement but the thing is it still just wholely makes up poo poo so the interface is useless i know alignment and related incorrect responses is a serious issue here but i absolutely think there's loads of strides to be made
|
|
#
?
May 30, 2023 18:54
|
|
|
go play outside Skyler posted:i think llms and generative ai will significantly change the way we interact with computers in the next 5 years the same way the iphone put a smartphone in everyone's hands yeah only that i can believe that in 5 years it will be significantly worse in all the aspects you mention
|
|
#
?
May 30, 2023 18:57
|
|
|
go play outside Skyler posted:i think llms and generative ai will significantly change the way we interact with computers in the next 5 years the same way the iphone put a smartphone in everyone's hands i think that is vastly overestimating how effective chatgpt and such will be
|
|
#
?
May 30, 2023 18:58
|
|
|
mediaphage posted:i know alignment and related incorrect responses is a serious issue here but i absolutely think there's loads of strides to be made i don't see how it would be possible to validate the entire dataset as being factual. even pairing it with a search engine isn't going to work so long as it starts feeding on its own bullshit, and i'd bet heavily that's already occurred
|
|
#
?
May 30, 2023 19:05
|
|
|
i do hold out a lot of hope that assistive technologies (not just for computers, but also e.g. bureaucracies, actual assistance on your complex personal circumstances) will be enabled by llm-derived stuff though. needs a lot of work puppetering the models reliably, but i think it fairly doable given a good environment to do the work. though the way things are going day by day right now i am despairing about that environment. does not take a lot of interference as that kind of outcome is not a big money-maker. Cybernetic Vermin fucked around with this message at 19:12 on May 30, 2023 |
|
#
?
May 30, 2023 19:05
|
|
|
grandmas everywhere already love alexa, after all
|
|
#
?
May 30, 2023 19:06
|
|
|
go play outside Skyler posted:i think llms and generative ai will significantly change the way we interact with computers in the next 5 years the same way the iphone put a smartphone in everyone's hands LLMs can’t do this
|
|
#
?
May 30, 2023 19:07
|
|
|
Improbable Lobster posted:i think that is vastly overestimating how effective chatgpt and such will be they are legitimately good at natural language processing. that's way too narrow a field to drive billions in investment though, so we're seeing that unchained. it's in their interest to leave it open-ended (while also claiming they don't know what it's capable of,  it might blow us up) because people will just fill in the blanks themselves. that ties back into the existential threat bullshit hype too — if enough people claim it's possibly capable of doing that, maybe it can do your taxes or something too (do not do your taxes with a llm)
|
|
#
?
May 30, 2023 19:09
|
|
|
hobbesmaster posted:LLMs can’t do this
|
|
#
?
May 30, 2023 19:12
|
|
|
"the torment nexus might destroy the earth!" proclaim the exact people who would definitely allow it to do so to hit their quarterly numbers
|
|
#
?
May 30, 2023 19:14
|
|
|
ffs, IT'S NOT THE COMPUTER FROM STAR TREK
|
|
#
?
May 30, 2023 19:14
|
|
|
hobbesmaster posted:LLMs can’t do this fill out a template? that's one of the things they do extremely well if you don't care about the actual content
|
|
#
?
May 30, 2023 19:14
|
|
|
BMan posted:ffs, IT'S NOT THE COMPUTER FROM STAR TREK have you considered maybe it could be soon, and you could miss out on the next big thing if you dont invest today?
|
|
#
?
May 30, 2023 19:15
|
|
|
|
| # ? May 12, 2024 14:29 |
|
|
this is elizabeth "steve jobs" "liz" holmes' reporting date to federal prison, let's see if she actually shows up or if her hotel heir husband hired a convincing body double
|
|
#
?
May 30, 2023 19:15
|
|