AI—image/text/paperclip maximizer?

[kerbiloid]

Just another emotion recognition training set.

Spoiler

 

 

[DDE]

P.S. I guess I'm working through the frustration that my first embedding has turned out as noise. I think I've fudged up by using the non-default model, and I maybe shouldn't have stated the gender of the person in the file description (if the logic is that you should describe everything besides what you want the AI to learn).

Edited April 10, 2023 by DDE

[SunlitZelkova]

College student admits to public broadcaster that he/she uses ChatGPT to cheat

https://www3.nhk.or.jp/nhkworld/en/news/20230411_03/

[TheSaint]

Well, that didn't take long.

Mom warns of terrifying AI voice cloning scam that faked kidnapping

Just wait until they start hitting the senior citizens with these calls. They already get a lot of them with just a voice actor.

[kerbiloid]

50 minutes ago, TheSaint said:

Well, that didn't take long.

Mom warns of terrifying AI voice cloning scam that faked kidnapping

Just wait until they start hitting the senior citizens with these calls. They already get a lot of them with just a voice actor.

A usual situation here without AI, just not abou kidnapping.

A histerical, crying girl's voice "Mom! I got into road accident, I'm in police. Please, quickly put money on this phone to make bail!"
An adult voice "It's a <bank name> security.  We need to check and confirm your bank account. Please, tell your card number."
An adult voice "It's an Investigation Commitee. We need your assistance, please tell us...", and so on.

(Warning, can be understood by somebody as politics, but just a truth of life.)

Spoiler

For the last decade, 95% of calls are coming from the currently well-known adjacent territory, believe or not.
A half-year ago a familiar woman transferred several tens million RURs to such scammers.

A standard practice here is/was to ask a keyphrase "Whose is Crimea? Say loudly."
For reasons, it was inappropriate for the "girl", "bank clerk", "police officer" and other characters to answer "Russian" in presence of their colleagues, sitting around.
So, they were starting to flame, swear, and dropped the call.
Currently they mostly adapted, so the detection phrases are changing.

Welcome to the brave new world.

[DDE]

2 hours ago, TheSaint said:

Just wait until they start hitting the senior citizens with these calls. They already get a lot of them with just a voice actor.

Forget voice acting. The kind of stuff @kerbiloid writes about has resulted in the elderly engaging in arson and shootings at bank branches, all at the behest of "policemen" and "bank security officers". They've even evolved to a script of "Hello, your card was used to donate to the Armed Forces of [redacted], do exactly what I say or you're going away for a long time". In Europe, the scenario seems to generally involve Interpol.

Also, for now, they're not using AI image generation. This specimen of a fake Bank of Russia employee ID I've come across just uses a sample picture from a photo retouching site.

Spoiler

 

[kerbiloid]

15 minutes ago, DDE said:

Forget voice acting.

We've received the "road accident" calls from the same idiotess girl several tens times during last months, until they got tired.

Cousin was called by his "son" (he doesn't have a son). And so on.

Edited April 12, 2023 by kerbiloid

[TheSaint]

We were talking about it tonight at the bar (the ultimate round table) and the final conclusion was that the endgame for AI is that it will be the end of information technology. Nobody will trust anything related to computers anymore. Video or audio evidence in court? Obviously a deep fake. Cryptocurrency, or even electronic bank records? Obviously hacked by AI. Not that it's true, or even possible. But ignorant juries will totally buy it. If they can't put their hands on it, or hear it from the mouth of an eyewitness, it will be suspect. We're on our way back to the 19th Century.

[tater]

On 4/10/2023 at 1:28 PM, SunlitZelkova said:

College student admits to public broadcaster that he/she uses ChatGPT to cheat

https://www3.nhk.or.jp/nhkworld/en/news/20230411_03/

A friend of my daughter's in college is also from her pretty rigorous HS. She said that he told her he wrote whatever assignment (they wrote so much that banging out several pages on literally anything is a joke to them), but then dumped it into GPT and asked it to grade it and provide suggestions—which he then used as a guide to rewrite it.

They might honestly have to have written assignments with proctors, no devices, etc. Course in the real world everyone will now just cheat I guess. How will we judge people on their ability to write if it's all written for them? Course those of us with a lot written could train the AI on our own writing, then have it write in our voice. Fun times?

 

34 minutes ago, TheSaint said:

We were talking about it tonight at the bar (the ultimate round table) and the final conclusion was that the endgame for AI is that it will be the end of information technology. Nobody will trust anything related to computers anymore. Video or audio evidence in court? Obviously a deep fake. Cryptocurrency, or even electronic bank records? Obviously hacked by AI. Not that it's true, or even possible. But ignorant juries will totally buy it. If they can't put their hands on it, or hear it from the mouth of an eyewitness, it will be suspect. We're on our way back to the 19th Century.

There was part of a recent Neal Stephenson book where something like this happened. A character became involved with a woman (married, but not sure if they had married yet), and they were tech people, sorta famous. Anyway, there was some dirt on his wife that got spread on the net somehow—social media presumably. Anyway, the solution was to have bots write thousands of other scurrilous things about her, then it turned loose and wrote them (MANY) about everyone. People then ignored any such reporting, since it was 99.9999% fake.

I suppose real audio/video/social media will eventually need to be definitively tied to the author/hardwware with a crypto key. Then if there is a faked vid, at least the key can be checked.

Maybe we need a Butlerian Jihad?

Also:

Quote

In this paper, we presented an Intelligent Agent system capable of autonomously designing, planning, and executing complex scientific experiments. Our system demonstrates exceptional reasoning and experimental design capabilities, effectively addressing complex problems and generating high-quality code.

https://arxiv.org/pdf/2304.05332.pdf

[TheSaint]

12 minutes ago, tater said:

I suppose real audio/video/social media will eventually need to be definitively tied to the author/hardwware with a crypto key. Then if there is a faked vid, at least the key can be checked.

That's the problem. Crypto key vs AI hack vs crypto key vs AI hack vs crypto key etc etc etc. Eventually this is all going to all get so convoluted that the average person is going to just throw up their hands and say, "I'm not buying the BS anymore!" And that is when it's all going to start to fall apart. Go do some reading on the real reasons why hyperinflation happens. Hint: It has nothing to do with printing too much money. That happens after the fact.

12 minutes ago, tater said:

Maybe we need a Butlerian Jihad?

I've been saying it for years. Because you are all my friends. Stock up on green tip.

[tater]

5 minutes ago, TheSaint said:

That's the problem. Crypto key vs AI hack vs crypto key vs AI hack vs crypto key etc etc etc. Eventually this is all going to all get so convoluted that the average person is going to just throw up their hands and say, "I'm not buying the BS anymore!" And that is when it's all going to start to fall apart. Go do some reading on the real reasons why hyperinflation happens. Hint: It has nothing to do with printing too much money. That happens after the fact.

I suppose it would have to be built into devices. You have your phone, and two factor authentication. Raw footage/etc is stamped as created by your device at time/place.

Still, it's gonna be a mess. And soon, very, very soon.

[kerbiloid]

1 hour ago, tater said:

You have your phone, and two factor authentication.

And the wider it's used, the wider it's known.

Once several agents had identified you, the probability that somebody doesn't need to ask you for those IDs grows exponentially, because this secret info gets stored in many places.

2 hours ago, TheSaint said:

We were talking about it tonight at the bar (the ultimate round table) and the final conclusion was that the endgame for AI is that it will be the end of information technology. Nobody will trust anything related to computers anymore. Video or audio evidence in court? Obviously a deep fake. Cryptocurrency, or even electronic bank records? Obviously hacked by AI. Not that it's true, or even possible. But ignorant juries will totally buy it. If they can't put their hands on it, or hear it from the mouth of an eyewitness, it will be suspect. We're on our way back to the 19th Century.

In XIX century they were living in fog due to no available methods of fast travel or photo fact fixation.
They could just believe what they read or hear.

In XXI century the picture will get compromised by AI deep fakes, and be in fog.

In XX century they were in the middle, when the XIX fog was being replaced with XXI fog.
The fake photo (and since late 1990s - videos) were known all XX long, and the fact filtering and misinterpretation were as wide as always.

The authentication problem of finances and private property is the easiest.
Once AI had captured any moneysome job, there will be no problem of keeping one's absent money. A beggar can never be bankrupt.
Everything other will be properly depictured and explained by Matrix in Matrix terms, and you will even see nothing wrong, unless your stomach gets rioting for more synthesized organics.

1 hour ago, tater said:

A friend of my daughter's in college is also from her pretty rigorous HS. She said that he told her he wrote whatever assignment (they wrote so much that banging out several pages on literally anything is a joke to them), but then dumped it into GPT and asked it to grade it and provide suggestions—which he then used as a guide to rewrite it.

They might honestly have to have written assignments with proctors, no devices, etc. Course in the real world everyone will now just cheat I guess. How will we judge people on their ability to write if it's all written for them? Course those of us with a lot written could train the AI on our own writing, then have it write in our voice. Fun times?

Usual times. Google and wiki made school exams and university projects much easier to compile. Remote lessons (covid, I'm looking at you) have developed this significantly. AI will just polish it.

1 hour ago, tater said:

Course in the real world everyone will now just cheat I guess.

"Will"? These 'Muricans...
(I heard they even don't look in each other's workbook at school, and sit at separated desks to prevent that... Was sounding actually funny and weird for us, plain and simple Soviet pupils.)

Yes, devices are  prohibited. But the cams in toilet cabins are prohibited, too. And the school doesn't want to be known for poor results of its pupils. So, the living life is stronger than the dead rules.

Well, of course a fan of the trials subject needs no cheating. He needs brakes to fit the paper and time.

1 hour ago, tater said:

I suppose real audio/video/social media will eventually need to be definitively tied to the author/hardwware with a crypto key.

How do you know if the author itself exists in flesh?

1 hour ago, tater said:

Maybe we need a Butlerian Jihad?

AI will inform you that it's succeeded! People won! Just watch TV with your own eyes, everything is there!

1 hour ago, TheSaint said:

That's the problem. Crypto key vs AI hack vs crypto key vs AI hack vs crypto key etc etc etc. Eventually this is all going to all get so convoluted that the average person is going to just throw up their hands and say, "I'm not buying the BS anymore!"

"Buying"? Maybe you even want to say "for money"?
https://en.wikipedia.org/wiki/Trudoden

Don't worry, tovarishch. You will have enough listed civilian trudodens from community works to receive your civilian ration from the store.

1 hour ago, tater said:

I suppose it would have to be built into devices. You have your phone, and two factor authentication. Raw footage/etc is stamped as created by your device at time/place.

The only unique device you actually have is you. And its signature is still to be stored somewhere and requested from there without touching you.

Edited April 13, 2023 by kerbiloid

[DDE]

Well, ****

https://nypost.com/2023/04/12/ai-clones-teen-girls-voice-in-1m-kidnapping-scam/

[magnemoe]

18 hours ago, tater said:

There was part of a recent Neal Stephenson book where something like this happened. A character became involved with a woman (married, but not sure if they had married yet), and they were tech people, sorta famous. Anyway, there was some dirt on his wife that got spread on the net somehow—social media presumably. Anyway, the solution was to have bots write thousands of other scurrilous things about her, then it turned loose and wrote them (MANY) about everyone. People then ignored any such reporting, since it was 99.9999% fake.

I suppose real audio/video/social media will eventually need to be definitively tied to the author/hardwware with a crypto key. Then if there is a faked vid, at least the key can be checked.

Maybe we need a Butlerian Jihad?

Also:

https://arxiv.org/pdf/2304.05332.pdf

Thought of this years ago, getting an video of you in such an setting and have serious money. Dump hundreds of them, all with fantasy stuff and combinations like  Hobit and Na'vi, Khajiit and mermaid, dridder and Dranei all having fun. 
Think of it like internet jamming as in releasing decoys, and obviously doing stuff like this for lots of people makes it much more foolproof. 
Azure will have crap performance that weekend but you can always refund. 

Because teens are the main users of social media the smarter ones will not use anything who is legally binding. 
Its not the law who is their problem, its parents, other teens and the school.  
the circles did not line up so you created multiple stories for each audience. 
I assume this is very old,   Neanderthals teens had this but I guess it became serious with the upper classes of the first high cultures like ancient Egypt. 

Teens has always pioneered new  technology for their purpose, text messages on mobiles was nice as they could chat in class. 
(assume so as got out of university couple of years before) 

Facebook got taken over by the parents and its an awesome tool for local communities and niche groups like veteran tractors. 
But the teen had to leave it as they did not want to get into the system for good reasons. 

Glad I grew up before social media, even if it probably been safer and more lucrative after it. 
However knowing more about explosives and rockets is not always an good thing. 

[DDE]

Russia's Sber has rolled out a free platform of their own.

https://fusionbrain.ai/en/

[kerbiloid]

1 hour ago, DDE said:

Russia's Sber has rolled out a free platform of their own.

https://fusionbrain.ai/en/

They say, it's like the year-old MidJourney. Tested. Faces are ok, fingers are pathetic.

[tater]

https://arstechnica.com/information-technology/2023/04/surprising-things-happen-when-you-put-25-ai-agents-together-in-an-rpg-town/

Pretty cool.

 

And a quick, in browser Stable Diffusion:

 https://clipdrop.co/stable-diffusion

 

[TheSaint]

1 hour ago, tater said:

https://arstechnica.com/information-technology/2023/04/surprising-things-happen-when-you-put-25-ai-agents-together-in-an-rpg-town/

Pretty cool.

It's turtles, all the way down.

[kerbiloid]

7 hours ago, tater said:

https://arstechnica.com/information-technology/2023/04/surprising-things-happen-when-you-put-25-ai-agents-together-in-an-rpg-town/

Quote

A group of researchers at Stanford University and Google have created a miniature RPG-style virtual world similar to The Sims, where 25 characters

Sounds ominously...

Should we prepare for the future AI revenge?

More experiment ideas for the Stanford University.

https://fallout.fandom.com/wiki/List_of_known_Vaults

[TheSaint]

LOL

‘Overemployed’ Hustlers Exploit ChatGPT To Take On Even More Full-Time Jobs

Rushing headlong to the end of work as we know it.

[DDE]

1 hour ago, TheSaint said:

Rushing headlong to the end of work as we know it.

Thus spake Ludd.

This is hardly the end of work, but it does present an interesting turn of events for an economy that is, at this point, heavily focused on dishing out paperwork.

[TheSaint]

1 minute ago, DDE said:

Thus spake Ludd.

This is hardly the end of work, but it does present an interesting turn of events for an economy that is, at this point, heavily focused on dishing out paperwork.

You misquote me.

1 hour ago, TheSaint said:

the end of work as we know it.

There's been a lot of hay strewn about the large percentage of modern jobs that are simply busywork. (Mostly by an entire new class of people who earn a living writing about other people's jobs on the Internet. Roll that around in your gourd.) I think AI is going to wreak havoc with the income earning potential of wide swaths of the population.

[tater]

20 minutes ago, TheSaint said:

There's been a lot of hay strewn about the large percentage of modern jobs that are simply busywork. (Mostly by an entire new class of people who earn a living writing about other people's jobs on the Internet. Roll that around in your gourd.) I think AI is going to wreak havoc with the income earning potential of wide swaths of the population.

The existential risk enthusiasts (aka, "doomers") have a point at the widest possible view, but the disruption of current work by AI, even systems that are already here, is a known issue, and I agree, it's gonna get ugly. I'm not sure this is the same as cars replacing horse, etc.

[DDE]

[kerbiloid]

Once somebody will familiarize this AI with the language and arts...