[Azath Vitr (D'ivers]

Maark Abbott, on 12 January 2023 - 08:35 AM, said:

AI is nothing but parrots, trained to imitate.

That's probably a fair characterization of the AI Apple is using for these audiobooks (so far...).

But AI is capable of excelling at creativity. Chess is a great example, though obviously the possibilities are a bit restricted, and the goal is clearly defined. To the extent that creativity can be characterized as a search through a space of possibilities, and AI can accurately learn what that space is, AI has a major advantage over human creativity---which may be greatly sped up by quantum machine learning.

ML models can be like superparrots.

What about understanding? LLM like ChatGPT easily pass the Turing test; Google's LaMDA, being connected to the internet, can properly index visual images and code, though being limited to language it can't yet analyze those images (for that it would need to be integrated with other AI) or execute the code. The 'symbol grounding problem' may be resolved by allowing AI to interact with simulations of the world (or with the external world itself through robots, though readings from accurate simulations of the range of possible inputs could do as well or better).

They are of course fundamentally unlike parrots in that they can form novel sentences with correct syntax that answer questions they've never been presented with before correctly.

Between the rising parrots and deflated angels, the aping of the ape is lifted on sublimely vast entangled networks of wings (and cloaca)....

[Maark Abbott]

Unsurprisingly, the AI streamer that had been live for a few days was trained by the internet to engage in Holocaust denial.

But yeah this is 100% the future.

[Maark Abbott]

Entirely unsuirprisingly: Midjourney, Stability AI and DeviantArt are getting hit with a class action. It was inevitable really: throwing two copyrighted images into a theft generator and then claiming it's a new entity was only ever going to come to this.

[Azath Vitr (D'ivers]

Maark Abbott, on 16 January 2023 - 08:40 AM, said:

Entirely unsuirprisingly: Midjourney, Stability AI and DeviantArt are getting hit with a class action. It was inevitable really: throwing two copyrighted images into a theft generator and then claiming it's a new entity was only ever going to come to this.

The lawsuit claims the AI is storing compressed 'copies' of the three artists' work. That is factually false, though in cases of overfitting it may be close enough to be debatable. However one of the artists recently wrote an op-ed in the New York Times admitting that AI does not produce copies of her work, but only imitates her style (which incidentally is extraordinarily crude, and---as she also confesses---a melange of derivative elements she took from other artists; and she also explains that imitating her style even more exactly does not violate copyright). So unless they found examples of AI almost exactly reproducing one of their pieces---which seems extremely unlikely---this lawsuit is essentially frivolous.

Btw the first AI trial lawyers are about to be deployed in the United States:

'The glasses will record the proceedings and a chatbot[...] will offer legal arguments in real-time, which the defendants have pledged to repeat[...] The locations of the hearings have been kept secret, to prevent judges from derailing the stunts ahead of time.

[...]

“My goal is that the ordinary, average consumer never has to hire a lawyer again,” [...]

“Courts should allow it, because if people can’t afford lawyers, at least they can have some help.”'

An AI Lawyer Is About to Defend a Human in a U.S. Courtroom

[Azath Vitr (D'ivers]

'the suit claims that AI art models “store com­pressed copies of [copyright-protected] train­ing images” and then “recombine” them; functioning as “21st-cen­tury col­lage tool[s].” However, AI art models do not store images at all, but rather mathematical representations of patterns collected from these images. The software does not piece together bits of images in the form of a collage, either, but creates pictures from scratch based on these mathematical representations.'

AI art tools Stable Diffusion and Midjourney targeted with copyright lawsuit - The Verge

[Tsundoku]

Azath Vitr (D, on 16 January 2023 - 11:29 AM, said:

Btw the first AI trial lawyers are about to be deployed in the United States:

'The glasses will record the proceedings and a chatbot[...] will offer legal arguments in real-time, which the defendants have pledged to repeat[...] The locations of the hearings have been kept secret, to prevent judges from derailing the stunts ahead of time.

[...]

“My goal is that the ordinary, average consumer never has to hire a lawyer again,” [...]

“Courts should allow it, because if people can’t afford lawyers, at least they can have some help.”'

An AI Lawyer Is About to Defend a Human in a U.S. Courtroom

[Azath Vitr (D'ivers]

Tsundoku, on 16 January 2023 - 01:27 PM, said:

Azath Vitr (D, on 16 January 2023 - 11:29 AM, said:

Btw the first AI trial lawyers are about to be deployed in the United States:

'The glasses will record the proceedings and a chatbot[...] will offer legal arguments in real-time, which the defendants have pledged to repeat[...] The locations of the hearings have been kept secret, to prevent judges from derailing the stunts ahead of time.

[...]

"My goal is that the ordinary, average consumer never has to hire a lawyer again," [...]

"Courts should allow it, because if people can't afford lawyers, at least they can have some help."'

An AI Lawyer Is About to Defend a Human in a U.S. Courtroom

https://www.youtube....h?v=Cqk-CLxrW6s

Or: imagine a world where everyone can easily afford a robo-lawyer that is perfectly sociopathic and trained specifically on the judge they'll be arguing before, based on past similar cases featuring similar litigants.

Corporations, rich people, and abusive government agents may have a harder time bullying people into settling or pleading guilty or no contest.

OTOH more people would be able to file lawsuits... but might have to pay the other side's legal fees if they lose, which could make individuals reluctant to sue corporations, rich people, or the government.

US law still dictates that anyone giving out legal advice has to have a license. Even though the US Supreme Court has ruled that 'corporations are people',

'It is not a lawful business except for members of the bar who have complied with all the conditions required by statute and the rules of the courts. As these conditions cannot be performed by a corporation, it follows that the practice of law is not a lawful business for a corporation to engage in.'

Democratizing Legal Services: Obstacles and Opportunities

So unless the law changes, it will have to be administered by lawyers. But fewer lawyers and legal aides will be required. IDK if a single lawyer would be able to get away with remotely monitoring the real-time performance of AI for multiple trials simultaneously.

Wonder if they'll beat the world record for number of people suing them simultaneously....

This post has been edited by Azath Vitr (D'ivers: 16 January 2023 - 02:50 PM

[Tsundoku]

Damn, apparently this AI generated art - Star Wars as 80s anime style by some program called "midjourney"

[Tsundoku]

"Sarah Connor?"

https://www.betootaa...s-headquarters/

[Macros]

And in a development that is depressingly unsurprising, software that mimics celebrities by sampling their voice has been used by 4chan to give us Hermione reading Mein Kampf and much more.

I mean, where the harm of these kinds of aggregating software AI am I right?

[Maark Abbott]

It being used for harm is a given - of course, AI simps will still argue that these programs are merely learning and that some voices sound similar so it's fine, but I can see this being used 99% for harm and 1% for Dagoth Ur shitposting.

On the subject of which


https://cdn.discorda...0-38-32-113.mp4

[Tsundoku]

More AI generated art - LOTR as an 80s anime. Not bad, actually.

[Maark Abbott]

Tsundoku, on 28 February 2023 - 09:25 AM, said:

More AI generated art - LOTR as an 80s anime. Not bad, actually.

https://www.youtube....h?v=isEWf9MCbJk

Whilst they have the capacity to be hilarious, AI voice programs are putting together extremely convincing audio clips that are borderline indistinguishable from the person they are approximating. The capacity for identity fraud is astonishing.

[Tsundoku]

It's real and it's happening NOW. We are done.

https://www.news.com...e7eb94ed6116ec6

How to use GPT-4: Image shows the world has changed forever
GPT-4 is already changing the world — allowing people to easily make websites, computer games and even businesses.

[Tsundoku]

‘Everything to lose’: Inside AI’s terrifying future as bots tell husband to leave wife
It has already broken up a marriage and saw a man take his own life. As the world embraces the new technology, this is the dark side of artificial intelligence.

https://www.news.com...565f6cda8e861a5

So, conceivably ChatGPT v10 (or even earlier) could deepfake video and audio to show on camera you committing a crime. I have also heard there are DNA kits that can mass produce traces of DNA from a small sample, that could be used to say, spread all over a crime scene. ChatGPT then checks your schedule to see when you have a window to commit a crime, with no witnesses and noone to provide you with an alibi. It then manipulates GPS data to show your phone and/or car at a scene of a crime. See where this is going? How long before it gets done? Has it been done already? Should this be done in a doco to demonstrate how easy it is?

How will we be able to prove we DIDN'T do it?

Then throw in the dark side of the health tracing benefits - AI-designed viruses. Maybe even targetting specific DNA.

How do you restrict this? I don't believe you can. All it takes is one group somewhere to let it run wild and Skynet becomes just one of a bunch of worries.

Sure, it makes a lot of things easier, but the downside scares the shit out of me.

This post has been edited by Tsundoku: 11 April 2023 - 12:15 AM

[Aptorian]

Obviously you just create a RoboCop AI to defeat the criminal AI. And then have a lawyer AI represent you in AI court, to get the AI judge to clean your record..

I was having a talk with a couple of colleagues at the library recently and I don't think they really understand how much of an impact Chatbots will make to the library trade. There's already a website that can recommend you library books based on a set list of criteria, like theme, genre, setting, language difficulty, pacing, etc. Combine this with our database of book reviews and you've basically eliminated 99% of the need for a professional librarian. As the bots become more humanlike the front stage purpose of our staff can be handled by minimum wage hosts that point you to the toilets and the elevator, and even that can be replaced by a tablet screen on wheels.

Of course, the constructive approach will be to have the staff work along with the AI. Assisting one another, learning from each other. But it's still daunting for our older staff who will seriously need to Pivot in order to stay useful

This post has been edited by Aptorian: 11 April 2023 - 01:01 PM

[Tsundoku]

Hollywood strikes inflamed by claim AI could do writers' jobs

https://www.news.com...7c8a07bb25328cb

-----------------

Not sure about this. AI creativity or originality? Maybe if it pumps out hundreds or thousands of random plots you might find some gold ... but who will find it and make that decision?
Will it then be that it creates a new class of editor whose job is purely to sort the wheat from the chaff?
Can AI do original or topical jokes? Sarcasm? Satire?

Be worried about when the AI can do those things consistently.
Of course that might also mean we are too close to becoming robots outselves with our preferences.

This post has been edited by Tsundoku: 07 May 2023 - 02:02 AM

[Tsundoku]

Dark cloud over ChatGPT revolution: the cost

https://www.news.com...c446c3cb4e3b484

------------------------------------

Hadn't really thought it would cost THAT much, but it's not exactly my forte.

Maybe when they build these big server farms (and in the current ones) they could also install heaps of explosives. Just in case.

[Macros]

 366389023_674392334734760_5098956289938988273_n.jpg (129.67K)
Number of downloads: 3

[QuickTidal]

Well that's fucked up.