AI is just a plagiarism/IP theft machine with extra steps which will used by the intellectually and creatively devoid 1% to steal the ideas of the lower classes and justify keeping us forever oppressed into manual labor.
Both what he was whistleblowing, but also his death, won't change anything about what's going to happen to generative AI (OpenAI's main 'product').
OpenAI isn't going to fail because what they're doing is illegal; it's going to fail because their industry and products are _dogshit_. They're panicking now because billions of dollars have been pumped into them, and the pressure to perform is getting bigger and bigger, but pressure can't make them NOT dogshit. Killing whistleblowers won't help.
Just like Boeing whistleblowers John Barrett and Joshua Dean, this is the oligarchy shoving it in the face of the people that they can kill us like how Russia kills dissenters, the United States and other powerful entities kill as well, fuck the oligarchy
So, this is what I was thinking about the other day. And it only applies to AI training for art and music purposes. What if they kept it to the following two groups of people. One, anything that is in the public domain (i.e. aged out of copyright) and two, those who willingly sign a document (hopefully with a big fat check) to allow their work to be included. That would seem to solve most of the concerns and still get good AI training.
I favour the russian one, man shot 4 times, in the chest, with 2 different calibre weapons, was found with a third calibre weapon in his hand... Suicide...
I will say, I feel like this kind of thing could be a bit insensitive if he really was going through some serious stuff, had other sincere problems that needed to be addressed, and instead everyone just uses his very personal decision as a pawn in a broader culture war.
Why would Russia kill someone who has information that could damage one of the most important companies in the United States, and why would the executives of that company be disinclined to call the hit?
If you're a witness in a corporate malfeasance case by default you should get or request witness protection. Fair to say organisations now operating as an out in the open mafia.
now, the question becomes who put the hit on him? was it the management at OpenAI or was the the AI?
Colossus: the Forbin Project eventually showed the machine putting out a hit on one of the developers that it considered superfluous. life imitates art.
What is this, a Michael Moore film? All he said was OpenAi was violating copyright. Not exactly alone on that one, and it's not like he specializes in copyright law. And this screenshot does indicate whether the death appeared violent. Why not be appalled when there is more information?
Basically he likely had a list of which IPs are being fed into it to train - much like how Midjourney had an explicit list of IPs and Artists to train, and they got nailed for infringement.
This should be pretty obvious to anyone who knows anything about generative LLM AI and does not really need any whistle-blowers. A friend of mine teachers this in the uni, specialised in copyright law.
They will never, ever be held accountable for this. Generative AI effectively functions by mixing the training data into one noise pattern and then acting like a de-noiser. If they're ever taken to court by a copyright holder, they'll simply stall the lawsuit out by arguing that this doesn't count as copyright infringement until it goes away. It's not even the only Chewbacca Defense they can throw at it, they can mix in others.
Yup. AI is a giant copyright infringement scam where they are going to use it to steal everyone's ideas so they don't have to pay them for any new creations.
Billionaires COULD give all their money away to people who need it more. Republicans COULD educate themselves on how reality works. CEOs COULD stop leeching the wealth produced by their labourers.
This may be an ignorant question, but isn't this an argument that any intellectual property lawyer could make about their software?
Or does it count as whistleblowing, because he's confirming that it's being trained on copyrighted materials? (Even then - the cat's kinda out of the bag now, isn't it?)
No, he didn't reveal any new information. He didn't say "I saw copyrighted material pass by my desk every day and I was supposed to pretend I didn't see it." You can still read his post: https://suchir.net/fair_use.html It's all just based on logic and understanding of how AI models work. Anyone could've "revealed" what he did.
OpenAI themselves confirmed that it's trained on copyrighted materials, in their statement before the House of Lords in the UK. (This is not an admission of any guilt; the fact is that everything created by everyone is automatically copyrighted, and it would be possible to obtain permission from the holders and train on those still-copyrighted materials.)
That isn't quite the same thing. That is obviously ripped off but could be fair use depending on the website. The AI was fed the query as well as the content of the top results as input to get one output.
But if the AI is trained on the data, the ripping off is not as obvious and yet that data is a permanent part of the neural network that is being used for every request from the AI, not just the one Google search. That should be illegal but the whistleblower was witness to non-obvious ripoff.
the thing is there is a diffrent from copying Grandmas cooking recpies blogspot site and you know copying all the data on a closed databse (that state that you can copy the data/information/text) and that in order to acsess it you need to pay a hefty subscription. OpenAI I just pay 5 usd for a one day pass and then copy everything and just cram it into the OpenAI database (literally stated not allowed yet they did it anyway).
Actually the new one is stating a fact in the "AI" generated precis, providing a "source" link from what you know to be a trusted site, but the precis and the source site don't match. Plus the whole thing is fucked anyway. Ask Alexa who the lead singer of "the cure" is, and it confidently says "buddy Holly".
Google's AI-powered results are made from a prompt like "do a Google search for whatever the user types and summarize the results." It is immediately being fed whatever information is in those results so it can summarize it. It doesn't mean the LLM was already trained on all of those results and is regurgitating it. Incidentally, this is why it's so bad, because it's drawing its answers from random posts online which can easily be wrong or jokes.
Yes these AI results are very concerning to me. Even though the garbage results from Google were already bought and paid for you at least had a choice and some ability to vet the information. How long before the AI result is all we get?
I mean, that's all true, but how exactly was he a "whistleblower"...? It appears that he just said the exact same thing that literally everyone else on the planet opposed to Generative AI has also been arguing.
It's not JUST "everyone else", the whole "is it fair use or not" is currently being looked over in many different courts world-wide. This specific argument was already brought up in those cases multiple times, so it's not like his talking to the press revealed anything that wasn't already included in legal battles.
Whistleblowers generally are not pulling back the curtain on some shocking revelations, rather, they expose insider knowledge that the criminal activity everyone knows is going on - but can't prove - is in fact going on. A bit like when law enforcement gets people in organized crime to testify. Everybody knows who the made men and mafia bosses are, but connecting them to crimes is the hardest part.
I mean, yes, but my understanding is that no one needed to be 'connected' to anything here. Not only was the company NOT trying to hide their activity, they were stating in court that yes, they obviously DID it, but they simply didn't think what they did should even be a crime to begin with.
Clearly, Suchir didn't agree with his bosses on that stance, but I'm not sure if his public disagreement makes him a "whistleblower." Broke an NDA, I'm sure, but "exposing" crimes? Not sure there.
How it works is that OpenAI goes "I am building this model for research", they contact google and every other data-hoarding company, and request the hose. They get an absurd amount of data, and they train their model.
Then, they go "model is good, here's research paper about it", and gives themselves the model for commercial use.
It's one thing to say "they're scraping the entire internet for training data" and quite another thing to say "I have this proof that they've been using copyrighted material from these Big Companies who will probably sue them out of existence for not being paid for permission beforehand." If he had evidence that any newspaper's content was used without permission, OpenAI would be done. Their entire business model at this point is restricting access to the fluff pieces. To say nothing of any >
> major publishers. Just imagine how much of a stink the Tolkein family would raise if they took in The Lord of the Rings without permission - then multiply that by thousands if Houghton Mifflin's works got scraped. Multiple entire industries could come baying for blood if there's even a shred of proof they were taken from without compensation. The entire future of LLMs could be at stake if he had the right proof. and given the fad they are, that's a lot of rich peoples' money to threaten.
No, he didn't reveal any new information. He didn't say "I saw copyrighted material pass by my desk every day and I was supposed to pretend I didn't see it." You can still read his post: https://suchir.net/fair_use.html It's all just based on logic and understanding of how AI models work. Anyone could've "revealed" what he did.
easyoneundone
hey, its the future
SecretlyASquirrel
Classic arms tied double shot to the back of the head suicide.
GreatOdinsTaint
AI is just a plagiarism/IP theft machine with extra steps which will used by the intellectually and creatively devoid 1% to steal the ideas of the lower classes and justify keeping us forever oppressed into manual labor.
secretagentchewbacca
Boksha
Both what he was whistleblowing, but also his death, won't change anything about what's going to happen to generative AI (OpenAI's main 'product').
OpenAI isn't going to fail because what they're doing is illegal; it's going to fail because their industry and products are _dogshit_. They're panicking now because billions of dollars have been pumped into them, and the pressure to perform is getting bigger and bigger, but pressure can't make them NOT dogshit. Killing whistleblowers won't help.
whowatcheswe
Just like Boeing whistleblowers John Barrett and Joshua Dean, this is the oligarchy shoving it in the face of the people that they can kill us like how Russia kills dissenters, the United States and other powerful entities kill as well, fuck the oligarchy
lassnerb12
I’m sure the police can get this solved in about a week…. They did it in NY.
wazeewa
Nah, he isn't a rich CEO so they will drag their feet then rule it a suicide
Roqinn
So he Epsteined himself?
drinkthederpentine
This is pure conspiracy theory bait without cause of death....
J0NKER
https://imgur.com/evAjQ6T
Zeplin
It was a suicide, and absolutely nothing points to anything different. Stop spreading conspiracy theories.
thebonesofmyancestors
So, this is what I was thinking about the other day. And it only applies to AI training for art and music purposes. What if they kept it to the following two groups of people. One, anything that is in the public domain (i.e. aged out of copyright) and two, those who willingly sign a document (hopefully with a big fat check) to allow their work to be included. That would seem to solve most of the concerns and still get good AI training.
tarquinious
EllisTomago
He probably tied his own wrists behind his back with an extension cord and then somehow sliced his own throat. Suicide.
CelestialSea
No, he just shot himself in the back sixteen times with a rifle.
waznewz
I favour the russian one, man shot 4 times, in the chest, with 2 different calibre weapons, was found with a third calibre weapon in his hand... Suicide...
unclesporky
I will say, I feel like this kind of thing could be a bit insensitive if he really was going through some serious stuff, had other sincere problems that needed to be addressed, and instead everyone just uses his very personal decision as a pawn in a broader culture war.
dcjunkee
Fell down an elevator shaft onto some bullets
PerroMalcriado
16 gunshots to the back, classic suicide
downvoteacct
Surprised he didn’t accidentally fall out of a window based on the direction this country is headed.
woozle
I'm sure it's just a coincidence
sezchuanllama
MassiveCollisionY
Maybe he accidentally tripped, fell very unfortunately into a bed with a pillow on top with a gun taped to the ceiling. It happens sometimes
quadraspaz1
...during wellbeing checks
Shaows
Boeing is subcontracting to make up lost funds
NewtGinGrinch
Russia killed whistleblowers, not Boeing execs. It's a "funny" joke but get a grip.
SerialChickenLover
Why would Russia kill someone who has information that could damage one of the most important companies in the United States, and why would the executives of that company be disinclined to call the hit?
Shaows
Sounds like something a Boeing plant would say.
SpectralSpider
So surely we're going to have a nationwide manhunt that dominates every fucking news cycle, right?
geotard
If you're a witness in a corporate malfeasance case by default you should get or request witness protection. Fair to say organisations now operating as an out in the open mafia.
Svartsinn
I googled "never pour acid onto water" and google AI told me yes (this is wrong)
Holte
Shiznit007
I wonder if anybody has heard of Pirate Bay
jimofoz
Was Alexa still on?
MarvelousMind
Corpos
phordlesscone
Goddamn, Boeing can't keep getting away with this!
taez555
Do people in Russia have the same reaction when people fall out windows?
NewtGinGrinch
Kremlin kills with poisons, plain clothes secret police, starvation in gulags..not just windows. You wanna be shocked? Go look at their
firlefranz
NewtGinGrinch
social media after an assassination/mysterious murder. The pro-death/dismissive shill spin-machine works them over the same way.
sandstruk
And stairs. Don’t forget the stairs
mikeatike
electric cables.
psmith00
now, the question becomes who put the hit on him? was it the management at OpenAI or was the the AI?
Colossus: the Forbin Project eventually showed the machine putting out a hit on one of the developers that it considered superfluous. life imitates art.
ThingsThatDontJustifyGenocide
Well the AI isn't capable of any such function, so a human did it, if it was a hit.
4chanrefugee
First victim of Roko's basilisk's...
ThingsThatDontJustifyGenocide
username accurate, geezus. (This is not a compliment)
HypnagogicHallucinations
What is this, a Michael Moore film? All he said was OpenAi was violating copyright. Not exactly alone on that one, and it's not like he specializes in copyright law. And this screenshot does indicate whether the death appeared violent. Why not be appalled when there is more information?
RigelKent
That's not what we do here. We must jump to conclusions.
rshini
Basically he likely had a list of which IPs are being fed into it to train - much like how Midjourney had an explicit list of IPs and Artists to train, and they got nailed for infringement.
supergreenpasta
Police said it looks like suicide, which would be what people expect if it's not a suicide. It's the timing that's suspect.
Revicus
Timing relative to what/when exactly?
supergreenpasta
His OpenAI whistle blowing, OpenAI's end of year announcements- lots of money invested into it.
fformulaa
This is what he was whistleblowing about
JeezusJuice
A class action over this was already dismissed.
thedadinblack4
Thank you for sharing this.
Shaodyn
One more time for the people in the back: "This is not a sustainable model for the internet ecosystem as a whole."
WheelieNelson420
This should be pretty obvious to anyone who knows anything about generative LLM AI and does not really need any whistle-blowers. A friend of mine teachers this in the uni, specialised in copyright law.
livingonagiantfireball
Lets put a contract on open Ai ceo head, pass the message to other ceo, maybe Boeing too
ssjalhazred
They will never, ever be held accountable for this. Generative AI effectively functions by mixing the training data into one noise pattern and then acting like a de-noiser. If they're ever taken to court by a copyright holder, they'll simply stall the lawsuit out by arguing that this doesn't count as copyright infringement until it goes away. It's not even the only Chewbacca Defense they can throw at it, they can mix in others.
DrFlukeHawkins
Yup. AI is a giant copyright infringement scam where they are going to use it to steal everyone's ideas so they don't have to pay them for any new creations.
ShrokNil
I mean one could make an AI that doesn't do that. The problem isn't AI it's how the AI is being built.
CanadianLadyMoose
Billionaires COULD give all their money away to people who need it more. Republicans COULD educate themselves on how reality works. CEOs COULD stop leeching the wealth produced by their labourers.
?1
Alas…
DrFlukeHawkins
And I get that it's the only way probably to do it...but then that means something needs to be done to compensate the people being stolen from.
Same people get mad if I sample 2 seconds of a song are just blatant thieves.
ButtersTheCat
This may be an ignorant question, but isn't this an argument that any intellectual property lawyer could make about their software?
Or does it count as whistleblowing, because he's confirming that it's being trained on copyrighted materials? (Even then - the cat's kinda out of the bag now, isn't it?)
rshini
the Latter, as Midjourney got busted because their CEO stupidly showed a list of IPs and artists they were explicitly training it on.
unclesporky
No, he didn't reveal any new information. He didn't say "I saw copyrighted material pass by my desk every day and I was supposed to pretend I didn't see it." You can still read his post: https://suchir.net/fair_use.html It's all just based on logic and understanding of how AI models work. Anyone could've "revealed" what he did.
unclesporky
OpenAI themselves confirmed that it's trained on copyrighted materials, in their statement before the House of Lords in the UK. (This is not an admission of any guilt; the fact is that everything created by everyone is automatically copyrighted, and it would be possible to obtain permission from the holders and train on those still-copyrighted materials.)
Suchir did not reveal any new information, just posted it in an informed and digestible way. https://suchir.net/fair_use.html
lightfoot2
He's right. If you look at Google or Bing's "AI" answer it always matches the top hit answer word for word. So profit for Google
airwoman
That isn't quite the same thing. That is obviously ripped off but could be fair use depending on the website. The AI was fed the query as well as the content of the top results as input to get one output.
But if the AI is trained on the data, the ripping off is not as obvious and yet that data is a permanent part of the neural network that is being used for every request from the AI, not just the one Google search. That should be illegal but the whistleblower was witness to non-obvious ripoff.
ZackWester
the thing is there is a diffrent from copying Grandmas cooking recpies blogspot site and you know copying all the data on a closed databse (that state that you can copy the data/information/text) and that in order to acsess it you need to pay a hefty subscription. OpenAI I just pay 5 usd for a one day pass and then copy everything and just cram it into the OpenAI database (literally stated not allowed yet they did it anyway).
Evi1Gav
Actually the new one is stating a fact in the "AI" generated precis, providing a "source" link from what you know to be a trusted site, but the precis and the source site don't match. Plus the whole thing is fucked anyway. Ask Alexa who the lead singer of "the cure" is, and it confidently says "buddy Holly".
astrangehop
My favorite is the horse that went to Mars in 1997 but nobody knows it's name
CJAW
That's bullshit, everyone knows Buddy Holly is the lead singer of Weezer
unclesporky
Google's AI-powered results are made from a prompt like "do a Google search for whatever the user types and summarize the results." It is immediately being fed whatever information is in those results so it can summarize it. It doesn't mean the LLM was already trained on all of those results and is regurgitating it. Incidentally, this is why it's so bad, because it's drawing its answers from random posts online which can easily be wrong or jokes.
fitgirl1
Yes these AI results are very concerning to me. Even though the garbage results from Google were already bought and paid for you at least had a choice and some ability to vet the information. How long before the AI result is all we get?
SteveMND
I mean, that's all true, but how exactly was he a "whistleblower"...? It appears that he just said the exact same thing that literally everyone else on the planet opposed to Generative AI has also been arguing.
igglebotato
He was inside the machine, I guess
ravnicrasol
It's not JUST "everyone else", the whole "is it fair use or not" is currently being looked over in many different courts world-wide. This specific argument was already brought up in those cases multiple times, so it's not like his talking to the press revealed anything that wasn't already included in legal battles.
mizzamir
Whistleblowers generally are not pulling back the curtain on some shocking revelations, rather, they expose insider knowledge that the criminal activity everyone knows is going on - but can't prove - is in fact going on. A bit like when law enforcement gets people in organized crime to testify. Everybody knows who the made men and mafia bosses are, but connecting them to crimes is the hardest part.
VodkaReindeer
Which industry are you thinking of? Because your description matches none of the whistleblowers I've heard about.
SteveMND
I mean, yes, but my understanding is that no one needed to be 'connected' to anything here. Not only was the company NOT trying to hide their activity, they were stating in court that yes, they obviously DID it, but they simply didn't think what they did should even be a crime to begin with.
Clearly, Suchir didn't agree with his bosses on that stance, but I'm not sure if his public disagreement makes him a "whistleblower." Broke an NDA, I'm sure, but "exposing" crimes? Not sure there.
rshini
I think is because he's an insider, he has explicit proof they're designed to lift things wholesale.
ravnicrasol
How it works is that OpenAI goes "I am building this model for research", they contact google and every other data-hoarding company, and request the hose. They get an absurd amount of data, and they train their model.
Then, they go "model is good, here's research paper about it", and gives themselves the model for commercial use.
Navrodel
It's one thing to say "they're scraping the entire internet for training data" and quite another thing to say "I have this proof that they've been using copyrighted material from these Big Companies who will probably sue them out of existence for not being paid for permission beforehand."
If he had evidence that any newspaper's content was used without permission, OpenAI would be done. Their entire business model at this point is restricting access to the fluff pieces. To say nothing of any >
Navrodel
> major publishers. Just imagine how much of a stink the Tolkein family would raise if they took in The Lord of the Rings without permission - then multiply that by thousands if Houghton Mifflin's works got scraped.
Multiple entire industries could come baying for blood if there's even a shred of proof they were taken from without compensation.
The entire future of LLMs could be at stake if he had the right proof. and given the fad they are, that's a lot of rich peoples' money to threaten.
unclesporky
No, he didn't reveal any new information. He didn't say "I saw copyrighted material pass by my desk every day and I was supposed to pretend I didn't see it." You can still read his post: https://suchir.net/fair_use.html It's all just based on logic and understanding of how AI models work. Anyone could've "revealed" what he did.