Maybe there is a way to fix the AI problem!? (Let's talk about how we can handle this sh*t)

[ghettosynth]

^^got this today in my feed. Also, Suno has a great Black Friday deal, just sayin.

[Touch The Universe]

The night is far spent. The prophecy of the image of the beasts unfolds upon a slumbering world. Creation awaits the dawn to arise and make it anew.

[Bunny_boy]

Only because you are looking at it incorrectly. Throughout your story you assign it human motivations - it wasn't being "emphatic", it simply repeated what it had previously said because that's all it's programming allows it to do. When I've encountered those kinds of situations, I have informed co-pilot that it was continuing to give me the same answer, which was incorrect, at which point it just stops trying and we both move on.

When I was writing that post, I was really conscious I was humanising it. I think that in itself is, or will be, a problem with capitulation to AI-produced data.
But in the end, the data that CoPilot gave me was incorrect: I know it's incorrect, but the general public wouldn't. The fact that it also gave me general references would be convincing enough for many/most people - I doubt most people check a stated reference for veracity. I've definitely done it before.

AI doesn't have to be like a human to demonstrate intelligence. Dogs, for example, show remarkable intelligence but they can't tell you anything at all, despite what Lassie movies might suggest. Their intelligence is different from ours. So is AI's version of intelligence.

Yeah, they're not going to tell you that Old Man Jenkins is trapped down the well, but I'm sure they can communicate their emotions!

I saw a video the other day where a bunch of Chinese doctors (in disguise, to avoid possible embarasment) went up against AIs in patient diagnosis. Guess what? The AI got to the correct diagnosis in seconds, where the doctors took 15 minutes or more in each case. You don't get to make doctors look foolish without some serious intelligence.

The big difference is that AI has not ability to make intuitive leaps. Instead it is able to go through a million options in seconds and come up with one that works. It's a different way of getting to the same point we can. It's not less or more intelligent, it's just different. If you like, it's faster than we are, not smarter. As I've said previously, it can pass the MENSA entry test and score well above the human average in many other forms of IQ test. So unless IQ tests are an invalid way of assessing intelligence, then you simply have to accept that AIs have high intelligence (coupled with incredible computational speed). There is no other logical conclusion to arrive at.

Yeah, these are specialised medical AIs. The technology is going to help loads with early diagnosis of cancers and other serious conditions. If they could be distributed into the community it would be an immense help.
At no point in the near future would anyone rely solely on the AI's diagnosis - it will always be checked by a clinician for sign-off.
And, of course, all this information is based upon input from humans.
I think that's totally true in this instance - AI is simply faster, not smarter. So the situation might become the AI does the testing and clinician does the sign-off.

There was a story I read or heard recently about training and AI on x-rays to diagnose a certain disease or condition. However, each x-ray also showed a stent, or some other implant. The AI therefore identified the implant as the disease / condition. Obviously that is just teething trouble, but it shows that human input is essential.

[bermudagold]

I saw a video the other day where a bunch of Chinese doctors (in disguise, to avoid possible embarasment) went up against AIs in patient diagnosis. Guess what? The AI got to the correct diagnosis in seconds, where the doctors took 15 minutes or more in each case. You don't get to make doctors look foolish without some serious intelligence.

cause the general Populus knows nothing of mathematics and engineering so they get caught up in the sensationalist hyperbole black box being called "AI"...they weren't following expert systems, computer vision, genetic algorithms, neural networks for the last 30 yrs...heuristics and metaheuristics have been at play all around them...medical field has known for well over a decade that software makes better medical imaging diagnosis than humans...software made the correct diagnosis 80% of the time...human doctors 60%...and when presented with the same data unknowingly a short time later, human doctors would make a different diagnosis 25% of the time...software doesn't get tired at the end of a 12 hr shift in ER, have low blood sugar from missing lunch, or be emotionally off balance from the angry phone call he just received from his wife

metaheuristics "AI" has been operating in commercial airlines to prevent them flying into each other since 1981..what did people think airline autopilots were?...airline industry has been trying to reduce the legally required amount of pilots on board aircraft for decades...eons before anybody had conceived a tesla

the point of AI was to do the things humans are bad at better...freeing us up for newfound time to pursue the creative tasks...that's why the interest and amount of venture capital being thrown at AI in the arts is bewildering to the pioneers

[BONES]

But in the end, the data that CoPilot gave me was incorrect: I know it's incorrect, but the general public wouldn't.

I'm not sure if you've noticed but at least half the shit you find on the internet is wrong. People will tell you flat-out lies at the drop of a hat, how could something people make, that is trained on lies and falsehoods, be expected to do any better? It's the old adage "garbage in, garbage out". If you are taught the wrong thing, how do you determine that it's actually wrong? I mean, most people can't see religion for the steaming pile of bullshit that it is, can they?

The way to correct AI is snot to tell it it's wrong, it's to provide feedback to the company that made it so they know it's got an issue, just as you'd do with any other software product you were using. I mean, if Cubase wasn't loading a plugin correctly, you wouldn't tell Cubase about it, you'd tell Steinberg.

Even our experience with using AI to write our next album has a success rate of less than one in a hundred. i.e. 90% of the results it gives us aren't want we want. So we stick at it, refining our prompts, until it gives us the correct output. People need to have realistic expectations. Working with AI in this way is probably the hardest I've had to work at getting an album together in my life.

Part of that is that we are moving so quickly with it, we've got half an album's worth of songs in really good shape in a little over a month, but it's a shitload of work taking the AI's output and getting it all working in Studio One with softsynths and MIDI. But it's new and exciting and I am enjoying the process way more than doing things the same old ways we've been working for decades.

Yeah, they're not going to tell you that Old Man Jenkins is trapped down the well, but I'm sure they can communicate their emotions!

That's something AI is also very, very good at. There are some amazing AI vocal performances on YouTube that you would swear were a real person singing something that was deeply personal to them. The vox in our AI generated material are also the big driver of the aggression in the songs, just as it should be. I'm going to have to go really hard to match that with my own vox. When you think about that, it should tell you everything you need to knew about working with AI.

The big difference is that AI has not ability to make intuitive leaps.

Of course not, why would you expect it to be otherwise? It's a perfect illustration of the difference between us and AI, not of a lack of intelligence on one side or the other.

At the end of the day, AI is a tool. Use it for the things it does well and do the other things another way. To reject it out of hand makes no sense. Or, you know, don't use it and everyone will think we're f**king geniuses and you're not.

[koalaboy]

You can't fix the AI problem until you fix the Human problem. Good luck with that.

[chagzuki]

The vox in our AI generated material are also the big driver of the aggression in the songs, just as it should be. I'm going to have to go really hard to match that with my own vox. When you think about that, it should tell you everything you need to knew about working with AI.

What AI tools are you using to manipulate or generate vox, Bones? For a while I've been hoping I could find a tool to shift vox in a certain direction a little while preserving a significant amount of the source. I'm not interested in what I've seen of the first generation of AI vocal replacement plugins, since they seemed almost entirely focussed on allowing users to swap in a sort of generic karaoke vocalist (which is anathema to me). I'm interesting in being able to subtly shift the expression, the problem being that I'm thinking of that shift in terms of existing singers, and of course that would be a copyright violation. E.g. if I want my voice to sound a little more alien/harsh I'm imagining I want to shift it 10 or 15% towards something like a Gary Numan target.

[BONES]

We're using Tunee, which spits out entire songs. We have to do the stem separation to isolate the vocals, which works quite well. It still takes a lot of work to get what you want, though. Tunee seems more interested in getting the melody and feel right than in pronouncing words correctly. You sometimes get gibberish.

We used Eleven Labs to do vocals on a track a couple of years ago and it did a way better job that I'd have expected putting in tone and emphasis. Initially I was only interested in getting guide track I could use to do my own vocals but it did such a good job they we ended up putting the AI voice on the track. It was just spoken word, though, not singing. Tunee does a decent range of styles, all within the Electro-Industrial realm, from almost whispering to a full-on scream.

[dexrow]

My stuff will be more popular when someone does me in so I dont need something like ai that can be detected in my music... If I can write a couple of odd plugins using ai then great although it appears that it writes programs in rust and then doesn't provide all the files that are required to compile them it is very annoying and rust is very annoying.

[eLawnMust]

Like everything else that used to be 'free' & now you must tolerate ads before or during otherwise upgrade to 'premium' what do you think will happen to AI once everybody becomes 'dependent' on it?... Everything goes the 'ad-whore' route when it used to be NOT!!

A week ago we were expecting company who we had never seen nor had ever even been out this way before & we told them to NOT use GPS as it does not work out here, follow directions... But like always NOPE! Now this old gal just blew past our driveway even though my wife was waving & yelling 20 feet from the road as the GPS was apparently telling her she needed to go a helluva lot further, An hour later she shows up...

Back in 2012 I sold a riding lawnmower on craigslist & told the guy do NOT use GPS... He was several hours late then I get a call, He's STUCK up in the woods & can't get out, freaking out & I said how did that happen? He reluctantly stated "Ohhh, I used the GPS!"... An hour later he get's 'unstuck' & makes it here in ANOTHER hour...

In 2009 wife & I going into town noticed a guy in one of those old 'renault Le Cars' which was white but with mud halfway up & he had a pissed look so I slowed to help & he shouted "I'm gonna SUE google!!"... I asked what happened, He said I was up there in the woods but google said I had a clear path but there was a river I went through... "uhh, You didn't SEE the river beforehand?"... No, I was watching the screen!!

I cannot even begin to imagine the 'AI Stupidity'.... I hope somebody blows this entire planet to shit...

[VOODOO U]

Everything goes the 'ad-whore' route when it used to be NOT!!

That's complete bullshit.
[This reply is sponsored by Coca Cola.Now available in Classic and ZERO sugar.]

NOPE!

Good movie.

2009 wife & I going into town noticed a guy in one of those old 'renault Le Cars' which was white but with mud halfway up & he had a pissed look so I slowed to help & he shouted "I'm gonna SUE google!!"... I asked what happened, He said I was up there in the woods but google said I had a clear path but there was a river I went through... "uhh, You didn't SEE the river beforehand?"... No, I was watching the screen!!

You know the most shocking thing in this story is that someone was driving a Le Car in 2009.

I hope somebody blows this entire planet to shit...

Humans will as they.move to.Mars.

[BONES]

Not happening, ask Neil DeGrasse Tyson.

[Zeisner]

https://futurism.com/the-byte/mit-predi ... e-on-track

If the third world war isn't going to fix the AI problem, this will.

[BONES]

Very exciting, I hope I live long enough to see this!

[vurt]

Very exciting, I hope I live long enough to see this!

get your robot thingy to do the soundtrack to the fall of civilisation.