|
dr_rat posted:Oh okay, so it knows how to emulate an acid trip far better than pretty much anything else seen in tv or movies. porn it takes specialized software and acres of data to make a simple deepfake using existing porn. If I can just type "Scarlett Johanssen having sex with my fursona" into the prompt and it spits out a full, original porn of that, well... that's gonna be worth some money.
|
#
?
Mar 29, 2023 02:07
|
|
|
|
| # ? Jun 7, 2024 20:22 |
|
|
Wow, Will Smith sure has gotten sloppy with eating spaghetti.
|
|
#
?
Mar 29, 2023 02:13
|
|
|
Shrecknet posted:porn Oh, yep right. I somehow forgot that. I am not looking forward to what the internet makes of that, but um at least it fits the thread title I guess!
|
|
#
?
Mar 29, 2023 02:39
|
|
|
Shrecknet posted:porn And will likely get nuked from orbit for obvious reasons.
|
|
#
?
Mar 29, 2023 02:39
|
|
|
Honestly I think that thing’s interpretation of furry porn might be less horrifying than its interpretation of human porn. The uncanny valley effect is just that bad.
|
|
#
?
Mar 29, 2023 03:32
|
|
|
StratGoatCom posted:And will likely get nuked from orbit for obvious reasons. or in other words, https://www.youtube.com/watch?v=c_o8vYUU-jo&t=17s Shrecknet fucked around with this message at 03:54 on Mar 29, 2023 |
|
#
?
Mar 29, 2023 03:49
|
|
|
dr_rat posted:Oh okay, so it knows how to emulate an acid trip far better than pretty much anything else seen in tv or movies. Actually the way these ai systems work is an awful lot like an acid trip. They have a bunch of random signals that they try and recognize, which layers slight distortions that make it look more like a particular object as it shifts around. It repeats this until it's ability to recognize objects is convinced the pseudorandom blob in front of it is Sean Connery in a leprechaun outfit. It draws this and tells the world it met Sean Connery on a spirit journey.
|
|
#
?
Mar 29, 2023 14:36
|
|
|
https://futureoflife.org/open-letter/pause-giant-ai-experiments/ Signatures include some big names. I personally agree with the letter, the speed of these developments seem unsafe but I doubt openAI and Microsoft will be willing to lose the current lead they have. Mega Comrade fucked around with this message at 16:40 on Mar 29, 2023 |
|
#
?
Mar 29, 2023 16:19
|
|
|
nobody would do anything useful about regulating ai in a six month timeframe
|
|
#
?
Mar 29, 2023 16:22
|
|
|
Mega Comrade posted:https://futureoflife.org/open-letter/pause-giant-ai-experiments/ Hahaha Please stop predicting the next word for six months is kinda overblowing these AIs a bit. People hyperventilating on both sides about text prediction is pretty funny. The companies over promising and critics believe them and have an existential crisis.
|
|
#
?
Mar 29, 2023 16:47
|
|
|
You can put a pause on the work Microsoft and OpenAI is doing and regulate the systems run by US and EU corporations but the threat is the systems that will be controlled and run by Russia, China and organized crime and there will be no stopping or slowing that.
|
|
#
?
Mar 29, 2023 16:56
|
|
|
At least half of these people signed this because their own work has been completely leapfrogged by that tech.
|
|
#
?
Mar 29, 2023 16:59
|
|
|
Heck Yes! Loam! posted:Hahaha I think AI taking over the world is a little silly but LLM dramatically reducing the headcount at a lot of companies is going to happen. Owling Howl posted:You can put a pause on the work Microsoft and OpenAI is doing and regulate the systems run by US and EU corporations but the threat is the systems that will be controlled and run by Russia, China and organized crime and there will be no stopping or slowing that. Creating models greater than gpt4 takes huge amounts of resources. Probably only china can do that at the moment. Honestly China has a lot to lose over mas redundancies caused by AI and they also have a much tighter grip on what it's tech sector is doing, if China agreed to stop, it would have a much easier time doing it than the US would. Mega Comrade fucked around with this message at 17:05 on Mar 29, 2023 |
|
#
?
Mar 29, 2023 17:03
|
|
|
They had to stop/pause signatures because it turns out the page auto updated whenever it was signed and all it verified was that the email given was valid. “Xi Jinping” was a signatory before they wised up. “Adolf Bonerhitler” was probably a signatory too. It’s too bad they didn’t use LLM to predict obvious human behavior. Boris Galerkin fucked around with this message at 17:09 on Mar 29, 2023 |
|
#
?
Mar 29, 2023 17:06
|
|
|
Heck Yes! Loam! posted:Hahaha The signatures include AI experts. And I don't mean Elon Musk, I mean actual experts. Head researchers at deep mind, Stuart Russell, Max Tegmark. These are not idiots playing with chatGPT and thinking it's sentient, these are people who have been studying AI for most of their professional lives who are concerned that we need legislation before proceeding any further because the speed at which things are progressing is dangerous. Mega Comrade fucked around with this message at 17:25 on Mar 29, 2023 |
|
#
?
Mar 29, 2023 17:15
|
|
|
Shrecknet posted:by who? Porn has won every single fight others have picked with it, because people really, really like porn. There is no world that visa/mastercard and any legitimate payment processor is going to provide services for a website that generates deepfake porn based off of real people. This leaves the shady payment processors based in like Russia or cryptocurrency. Someone who’s savvy enough to convert cash to buttcoins just to buy illegal deepfake porn is probably a lost cause in regulation and the people who enter their credit card info into realvisapayment.ru are gonna get hosed with their identity being stolen eventually. Next to the payment processors, you’ve got domain registrars, servers, cdns, etc, all of who are also not going to go anywhere near illegal deepfake porn. Then you’ve got laws like revenge porn laws that already exist today. Not to mention CSAM related laws that will absolutely gently caress people over. You’re right, deepfake porn is probably going to be a thing if it isn’t already. But it’s going to be relegated to the dark corners of the internet much in the same way CSAM is, and if they can pass laws for those types of contents I’m sure they can regulate deepfake porn under similar frameworks.
|
|
#
?
Mar 29, 2023 17:22
|
|
|
https://twitter.com/guntrip/status/1640694869785030657 https://twitter.com/guntrip/status/1640695226728587264 I don't think the chatbots are gonna destroy us all quite yet, folks
|
|
#
?
Mar 29, 2023 17:28
|
|
|
Mega Comrade posted:The signatures include AI experts. And I don't mean Elon Musk, I mean actual experts. Head researchers at deep mind, Stuart Russell, Max Tegmark. Yeah, while there are a myriad of issues with this petition, the underlying concern is legit. The issue here is that these models have severely outperformed expectations in a number of remarkable ways. There are a large number of unanswered questions at play, and our ability at guesstimating technological horizons has been severely compromised for the time being. This is objectively not good, and has large repercussions as far as policy-making, resource allocation, etc... goes. They are advocating letting go of the gas (not putting the brakes) until we get our bearings back. The problem is not about what they can do, it's that we did not see it coming. The main issue I personally have is that it's incredibly unrealistic, but I can definitely appreciate raising the alarm on this. edit: The other issue I have is more of a technical one: Model power is not connected that directly to number of parameters, making it a bad metric to target. It's been known since the 70s that too many parameters almost always leads to overfitting, and recent research indicates that there is still a lot more performance to be extracted from LLMs with sub-GPT4 parameter counts. Also, while the people arguing about consciousness, the singularity, etc... are over the top silly, the "It's just predicting the next word" and the "It's just a very fancy search through the training data" framings are not much better. This may be how we interface with these systems, but that informs little about its capabilities. Applying White Box reasoning to models of that magnitude just does not make sense. Aramis fucked around with this message at 18:02 on Mar 29, 2023 |
|
#
?
Mar 29, 2023 17:29
|
|
|
The issues they raise are fair, but they are societal issues not technological ones (which I think they are also saying) and that is exactly why that letter will achieve absolutely nothing. I've done my share of activism that led to sweet gently caress all in my time so no hate
|
|
#
?
Mar 29, 2023 17:34
|
|
|
woke kaczynski posted:https://twitter.com/guntrip/status/1640694869785030657 milkshake recipe sucks. your suppose to cut open the pods and scrape tje vanilla stuff into the THING, not sure how edible the dried out skin pods are. PhazonLink fucked around with this message at 17:39 on Mar 29, 2023 |
|
#
?
Mar 29, 2023 17:36
|
|
|
Shrecknet posted:porn This is a particularly fun example. For things that don't actually exist, like anthropomorphic screaming hairy armadillos, where does the training data come from? It has to come from art. But it's not like there's just one way to anthropomorphize an animal, on top of variations in art style. And most of these models still struggle with fingers. I feel like including enough furry poo poo that a model can handle different degrees of furryness--let alone sparkledogs with stars in their fur or three different kinds of wings--is gonna produce exciting new ways to malfunction and I am here for it.
|
|
#
?
Mar 29, 2023 18:02
|
|
|
PhazonLink posted:milkshake recipe sucks. Good poem though
|
|
#
?
Mar 29, 2023 18:09
|
|
|
Mega Comrade posted:The signatures include AI experts. And I don't mean Elon Musk, I mean actual experts. Head researchers at deep mind, Stuart Russell, Max Tegmark. Anybody who signed their name to this can be dismissed as somebody who will not be contributing to the future of AI. They don’t understand how people work.
|
|
#
?
Mar 29, 2023 18:10
|
|
|
PhazonLink posted:milkshake recipe sucks.
|
|
#
?
Mar 29, 2023 18:28
|
|
|
Aramis posted:Model power is not connected that directly to number of parameters, making it a bad metric to target. It's been known since the 70s that too many parameters almost always leads to overfitting, and recent research indicates that there is still a lot more performance to be extracted from LLMs with sub-GPT4 parameter counts. For sure, but have you considered that we can point at two datapoints to get infinite funding instead
|
|
#
?
Mar 29, 2023 18:38
|
|
|
tk posted:Anybody who signed their name to this can be dismissed as somebody who will not be contributing to the future of AI. They don’t understand how people work.
|
|
#
?
Mar 29, 2023 18:39
|
|
|
pumpinglemma posted:What else can they reasonably do besides make noise via the press, though? Start a bombing campaign against GPU manufacturers? A lot of these folks have been working in this space for years if not decades. If having these guidelines in place is so important, and they can be put together in 6 months, I’m wondering what exactly they’ve been working on without these guidelines. I don’t see any ideas, outlines, or proposals about what needs to be done. No plan on how things would come together so quickly. All I see is a bunch of FUD.
|
|
#
?
Mar 29, 2023 21:04
|
|
|
Politicians generally have a habit of never taking anything seriously until the 11th hour.
|
|
#
?
Mar 29, 2023 21:14
|
|
|
Aramis posted:The issue here is that these models have severely outperformed expectations in a number of remarkable ways. There are a large number of unanswered questions at play, and our ability at guesstimating technological horizons has been severely compromised for the time being. This is objectively not good, and has large repercussions as far as policy-making, resource allocation, etc... goes. They are advocating letting go of the gas (not putting the brakes) until we get our bearings back. The problem is not about what they can do, it's that we did not see it coming.
|
|
#
?
Mar 29, 2023 21:51
|
|
|
Boris Galerkin posted:There is no world that visa/mastercard and any legitimate payment processor is going to provide services for a website that generates deepfake porn based off of real people. This leaves the shady payment processors based in like Russia or cryptocurrency. Someone who’s savvy enough to convert cash to buttcoins just to buy illegal deepfake porn is probably a lost cause in regulation and the people who enter their credit card info into realvisapayment.ru are gonna get hosed with their identity being stolen eventually. This poo poo honestly is the only thing AI gen is useful for; it's looking like the USCO is sticking with the view that AI generated poo poo is uncopyrightable and rumblings are that at least some big names are starting to update contracts to ensure this poo poo doesn't enter their chains. And before someone invokes The Mouse, if that doctrine changes, we have bigger problems because the copyright system just had a sewer outlet constructed in its living room and the whole thing is going to lurch to a halt under a tide of AI copyright trolling. StratGoatCom fucked around with this message at 00:40 on Mar 30, 2023 |
|
#
?
Mar 30, 2023 00:38
|
|
|
Mega Comrade posted:The signatures include AI experts. And I don't mean Elon Musk, I mean actual experts. Head researchers at deep mind, Stuart Russell, Max Tegmark. What's dangerous isn't the speed that things are progressing; the stuff that's already out there is quite dangerous enough. The likely intention of the letter is to draw attention away from the current harms that "AI" tools are already inflicting right now, by instead placing the focus on hypothetical future harms that a potential future "human-competitive intelligence" could inflict. Elon Musk is signing (and loudly hyping up) a letter complaining that OpenAI is developing AIs too fast and too irresponsibly, while at the same time he's got tens of thousands of "self-driving" cars on the roads killing people and wrecking poo poo, which he'll happily advertise as being powered by "AI". The letter warns about the threat that powerful "nonhuman minds" might someday lead to "loss of control of our civilization", but it doesn't have a single word to spare for the real harms that ChatGPT3 and even basic-rear end machine-learning poo poo are already inflicting on us.
|
|
#
?
Mar 30, 2023 00:59
|
|
|
Rogue AI Goddess posted:I have a modest proposal for drafting policy and estimating horizons: use AI assistance to cut down time from 6 months to 6 minutes.
|
|
#
?
Mar 30, 2023 01:05
|
|
|
Honestly the stated goal and most of the content of the letters petty useless. It has seemed to of had a very small effect of starting a bit more conversation about AI's and potential harms which is good though and hopefully that will be the main take away from the letter. Although I think it will probably just be completely forgotten about within two weeks. As far as a call to action, I think it was a really dumb one and not a great starting off point for conversation, but it did get headlines I guess so  .
|
|
#
?
Mar 30, 2023 01:08
|
|
|
Incidentally it turns out to be really hard to get chatGPT to opine on how it could be used to replace workers and it will try to talk about how it can enhance their work product instead and would never replace them it'll happily write termination letters because they have been replaced by chatGPT and denials that the termination letters were automatically generated and insincere though
|
|
#
?
Mar 30, 2023 01:18
|
|
|
FWIW in my experience, having tested this a few times for shits and giggles, the form letters that ChatGPT drafts are completely awful and would absolutely require human intervention to not embarrass yourself.
|
|
#
?
Mar 30, 2023 01:32
|
|
|
If and only if the person using them gave even the tiniest gently caress.
|
|
#
?
Mar 30, 2023 01:37
|
|
|
https://twitter.com/CBCNews/status/1641040123721658369 Someone's supposedly running scams against older parents with not-entirely convincing deepfakes of their adult children's voices. quote:When Donna Letto got a phone call one day in December, she says, she recognized the voice on the other end of the phone. I'm not really sure what convinced them that this is actually a deepfake rather than a close-enough voice filter, besides some AI expert saying that it's possible and also the victims' own embarrassment at nearly falling for it.
|
|
#
?
Mar 30, 2023 04:17
|
|
|
Saw this on LinkedIn but if you run a google search for "regenerate response" -chatgpt You’ll find things in the index that were copy/pasted from chatgpt but not edited because they grabbed the regenerate response CTA When I tried this yesterday I found an old lady’s obituary and got too sad to even put it into words.
|
|
#
?
Mar 30, 2023 05:16
|
|
|
eXXon posted:https://twitter.com/CBCNews/status/1641040123721658369 I think pretty soon deepfake is just going to be the general term for "my voice/face was impersonated" even for stuff like a static photoshopped image.
|
|
#
?
Mar 30, 2023 05:35
|
|
|
|
| # ? Jun 7, 2024 20:22 |
|
|
drat we passwords for Stranger Danger as kids and now we need passwords for deepfakes.
|
|
#
?
Mar 30, 2023 06:58
|
|
