AI Ai ai...

[woody]

Since a couple of months I have this brand spanking new Samsung phone that is supposedly loaded with all sorts of smart AI shite that will help me through the day. Every now and then this intelligence looks through at all my pictures and then offers me a 'story' that I  (supposedly) like.

Yesterday I got one named 'Curious kitties in Warmenhuizen'.

And indeed, these were all pictures of our cats. The first picture it offered, while playing a happy tune, was of a dead cat.  This was Artemis, who was hit and killed by a car two months ago. The second picture a closeup of the same cat, blood and all.

So while AI seems perfectly capable of finding the kitties among my pictures, it lacks the sense to realize that maybe a picture of a dead cat or two might really not be what I would like to look at again.

I think it will take quite some time before AI will outdo us humans. At least in the emotional aspects of life 

[Siwastaja]

Typical AI: it's the same year after year after year, total lack of even most primitive context understanding. The thing is, this hasn't got any better, no matter how much resources are being poured into AI, and how some types of results have massively improved (e.g., fidelity of generated pictures, correctness of language constructs). But those photorealistic spotless pictures still have 6 fingers, well-explained long answers still explain how 1+1 = 3, and nice happy cat stories still involve dead and bloody cats.

It's probably more difficult to get right than many expect, and current zero rate of improvement on this aspect does not predict any success in near future, but we'll see.

[voltsandjolts]

This was Artemis, who was hit and killed by a car two months ago.

   RIP Artemis              ...nice name "Greek goddess of the hunt"

[woody]

AKA Diana, or, as we used to call her also, Lady Di (before we knew she was to end up under a car, that is)

[BrokenYugo]

When I hear "AI" I autocorrect in my head with "marginally more effective lying chatbot" or "latest investment scam buzzword". Helps to set expectations.

Starting to look like the next phone might be a flip phone if they're all polluted with that crap now.

[Ranayna]

First of all, as trite as it may sound: Sorry for the loss of your pet, and sorry that you had to take that AI mess.

I suspect one big issue here is, that AI is hardly trained on images like that.
Public image hosts and image sites are in my experience heavily moderated against images like dead cats. Understandably so.
But of course that limits what is available for AI to "learn" from.

[SteveThackery]

I've read an argument that says AI-generated pictures and text are finding their way onto the Internet, and current or future AI products will unwittingly be trained on material already produced by AI. And round it goes again for the next generation and the next. Do you think this sounds credible?

And what would the effect be? It feels intuitive to expect a kind of positive feedback effect whereby the AIs become more and more "random" and "confused", with errors and misconceptions eventually swamping any useful output. But that's just my intuition, which comes from a place of complete ignorance of the technology.

[SiliconWizard]

And the nice part is that Samsung feels entitled to send all your photos (and probably other personal data) - possibly in some predigested form, if you're lucky and they cared a bit about privacy, but I don't hold my breath too much - to some remote server to give you all these nice new features. How does that feel?
 

[Bud]

AKA Diana, or, as we used to call her also, Lady Di (before we knew she was to end up under a car, that is)

Seems Ladies Di and cars do not play together well.

[tom66]

I've read an argument that says AI-generated pictures and text are finding their way onto the Internet, and current or future AI products will unwittingly be trained on material already produced by AI. And round it goes again for the next generation and the next. Do you think this sounds credible?

And what would the effect be? It feels intuitive to expect a kind of positive feedback effect whereby the AIs become more and more "random" and "confused", with errors and misconceptions eventually swamping any useful output. But that's just my intuition, which comes from a place of complete ignorance of the technology.

At the "Dog Fair" over the weekend there were two stalls there that obviously had AI-generated artwork printed on mugs and T-shirts. The giveaway was a cat with slit pupils in the wrong direction (horizontal), and the dog with an extra paw in the shadow.

So, the idea of AI training off itself doesn't sound too unlikely to me, given it's already displacing actual artists.

[coppercone2]

I think that you should notify the manufacturer. I think what happened to you is really sad and some people could get really upset by this sort of deranged occurrence.

[woody]

And the nice part is that Samsung feels entitled to send all your photos (and probably other personal data) - possibly in some predigested form, if you're lucky and they cared a bit about privacy, but I don't hold my breath too much - to some remote server to give you all these nice new features. How does that feel?
 

Yeah, there's that. Completely unclear is what gets sent to Samsung and what is processed locally, by Qualcomms 'AI engine'. For me I assume that for every AI feature (like photo editing, real time translating) my data is going back and forth. How does that make me feel? Not too good. Unfortunately the world has decided that privacy is of no importance, convenience rules and our data is the currency that pays for it all. I fight this where I can but I realize it is a lost battle.

[woody]

I've read an argument that says AI-generated pictures and text are finding their way onto the Internet, and current or future AI products will unwittingly be trained on material already produced by AI. And round it goes again for the next generation and the next. Do you think this sounds credible?

Certainly. An effect of that self-feeding might be that at a certain point nothing new is ever going to be added. The large push behind AI is saving money. Using AI we can generate content for a field without having any specific knowledge of it. This is much cheaper than paying a specialist to do the same.

So, for argument's sake, if I ask ChatGPT to make me the code for a blinking led using an 18F55Q43 with the led on output C1 using a CCS compiler, it answers with compile-able (and working) code. While in itself this is a remarkable feat, when this code gets used by me and eventually is fed back into the internet the next training round the same code is going to be hoovered up and probably will get a higher quality score. AI is not ever going to make changes to this code, like humans might. Meanwhile the specialists who might have been able to change the output for the better have all moved to other jobs. And new specialists are not trained as there is obviously no money to be made in the jobs that disappeared.

But what do I know, I am not an AI specialist 

[SiliconWizard]

The large push behind AI is saving money. Using AI we can generate content for a field without having any specific knowledge of it. This is much cheaper than paying a specialist to do the same.

Both saving money *and* circumventing intellectual property, making it eventually obsolete. This second point also has major implications beyond the "money".

[woody]

The large push behind AI is saving money. Using AI we can generate content for a field without having any specific knowledge of it. This is much cheaper than paying a specialist to do the same.

Both saving money *and* circumventing intellectual property, making it eventually obsolete. This second point also has major implications beyond the "money".

Good point. That will also end innovation. Who will invest time and money on an idea when you know it is immediately purloined by Ai?

[tom66]

The large push behind AI is saving money. Using AI we can generate content for a field without having any specific knowledge of it. This is much cheaper than paying a specialist to do the same.

Both saving money *and* circumventing intellectual property, making it eventually obsolete. This second point also has major implications beyond the "money".

Good point. That will also end innovation. Who will invest time and money on an idea when you know it is immediately purloined by Ai?

All depends on how courts see IP going forward wrt to AI.  There will be a number of important test cases coming up in the next few years.

[unseenninja]

... So, for argument's sake, if I ask ChatGPT to make me the code for a blinking led using an 18F55Q43 with the led on output C1 using a CCS compiler, it answers with compile-able (and working) code. While in itself this is a remarkable feat, ...

There's absolutely nothing remarkable about being able to regurgitate some code which someone else wrote to "blink LED on 18F55Q43". Simply using a search engine will deliver the same result without you feeding all your prompts to ChatGPT for them to use as they will.

[Ranayna]

The large push behind AI is saving money. Using AI we can generate content for a field without having any specific knowledge of it. This is much cheaper than paying a specialist to do the same.

Both saving money *and* circumventing intellectual property, making it eventually obsolete. This second point also has major implications beyond the "money".

Good point. That will also end innovation. Who will invest time and money on an idea when you know it is immediately purloined by Ai?

All depends on how courts see IP going forward wrt to AI.  There will be a number of important test cases coming up in the next few years.

Regarding the output, there already have been a few court cases, and if i haven't missed something significant, the gist is, that AI created content is not protected by copyright, even if significant manual work was done to refine the result.

[woody]

... So, for argument's sake, if I ask ChatGPT to make me the code for a blinking led using an 18F55Q43 with the led on output C1 using a CCS compiler, it answers with compile-able (and working) code. While in itself this is a remarkable feat, ...

There's absolutely nothing remarkable about being able to regurgitate some code which someone else wrote to "blink LED on 18F55Q43". Simply using a search engine will deliver the same result without you feeding all your prompts to ChatGPT for them to use as they will.

Really? Putting this question:

get me ccs c code to blink a led with a 1 hz frequency on an 18F55q43 with a clock of 16 MHz and the led connected to RD4

to ChatGPT got me 5 seconds later :

Code: [Select]

#include <18F55Q43.h>        // Include the appropriate device header file.
#use delay(clock=16000000)    // Specify the clock frequency (16 MHz).

void main() {
   // Set RD4 as an output pin.
   set_tris_d(0b11101111);    // Set RD4 as output (0 for output, 1 for input in TRISD).

   while(TRUE) {
      // Toggle the state of RD4.
      output_toggle(PIN_D4);   // Toggle LED connected to RD4.

      // Wait for 500 milliseconds.
      delay_ms(500);           // 500 ms delay for 1 Hz blink frequency (500 ms ON, 500 ms OFF).
   }
}
Including an explanation on how it arrived at some of the choices.

Putting the exact same question to google.com gave me a shitload of information on blinking leds on various microcontrollers at various frequencies,  but none of the answers were even close to what I asked.

YMMV, but I find that remarkable.

[tom66]

Absolutely.  For programming problems, I find myself using ChatGPT about 50% of the time.  It hasn't replaced actual programming, but it's really useful for finding problems and giving short solutions to problems.

One of the recent ones I've had is where I can't trace a templating or syntax error in C++ down, I give the problem over to ChatGPT.  It hasn't failed me yet.

[SiliconWizard]

... So, for argument's sake, if I ask ChatGPT to make me the code for a blinking led using an 18F55Q43 with the led on output C1 using a CCS compiler, it answers with compile-able (and working) code. While in itself this is a remarkable feat, ...

There's absolutely nothing remarkable about being able to regurgitate some code which someone else wrote to "blink LED on 18F55Q43". Simply using a search engine will deliver the same result without you feeding all your prompts to ChatGPT for them to use as they will.

True, and at least you'd have the source rather than using some code generated from existing stuff but that you can't know where it comes from.

What people like with ChatGPT "as a search engine" is that it (usually) provides much more focused answers than a list of results in most search engines, with varying relevance.
But using LLMs to make search engines better was, I think, one of the whole points initially. Oddly enough, the attempts, so far, have been either non-existent or pretty bad. Where is the "AI" when you use Google? Shouldn't Google search be much better these days than it was 20 years ago? Yet, it's actually become worse.

[mendip_discovery]

I do find the AI information loop rather funny and it's going to interesting as it gets more messed up.

I just wish that it wasnt a hyped thing that is being forced into everything, and it being the mangement buzzword. I do have concerns that some will see it as being able to get rid of my humans from thier employment as they can get AI to deal with customer complaints, writing descriptions, making content etc.

[magic]

Shouldn't Google search be much better these days than it was 20 years ago? Yet, it's actually become worse.

Search became worse because SEO learned how to game the new algorithms and because there are lusers who actually want those stupid answers like an endless list of offers to buy a product when you actually look for information about it. The lusers don't care about information, they just want to be shown something to buy. The Internet has also seen a vast increase in the quantity of stupid content. That's also because you can make money on stupid content, if only you put ads on it and promote it - which brings us back to SEO.

[rstofer]

Where is the "AI" when you use Google? Shouldn't Google search be much better these days than it was 20 years ago? Yet, it's actually become worse.

Lately, my Google searches have had the first response tagged "AI" and it is the more comprehensive reply.  The riff-raff responses follow.  At least Google points out that the response was generated by AI.

[AVGresponding]

Shouldn't Google search be much better these days than it was 20 years ago? Yet, it's actually become worse.

Search became worse because SEO learned how to game the new algorithms and because there are lusers who actually want those stupid answers like an endless list of offers to buy a product when you actually look for information about it. The lusers don't care about information, they just want to be shown something to buy. The Internet has also seen a vast increase in the quantity of stupid content. That's also because you can make money on stupid content, if only you put ads on it and promote it - which brings us back to SEO.

No, it's because Google takes money from people in order to promote their product/service by having it ranked higher in a search.