• Welcome to ASR. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Master AI (Artificial Intelligence) Discussion/News Thread

Anybody ever use Pinterest?

If you haven’t used it for a couple years and visit, be prepared for a shock: aside from the user interface making it impossible to view any image without all sorts of menu or shopping crap all over it, it’s been overrun with AI slop. The site is practically useless now.

The slippery slope to dead Internet theory has begun.
 
  • Like
Reactions: mpw
So that's the reason why the rear wall (behind the speaker) is called front wall (in front of the listener) in Your language, not?

Not really.

Except in special cases such as Amir cites, the front is the front wrt the listeners. An orchestra plays at the front of the concert hall. A band performs on stage in front of the audience. The back rows are, surprisingly at the back of the hall. And my stereo speakers are in front of me, a bit in front of the front wall (which is thus behind them).

It depends on the reference or subject. So CEA-2034 refers to the loudspeaker which is the test subject when defining the back wall as behind the speaker.

Because chatbots don’t reason or understand concepts etc, their probabilistic outputs can run in all directions, depending on everything from pre-training corpus to end-user prompts. Engineers otoh may have problems with language, context and perspective. A specific definition shouldn’t be applied outside scope, or to the wrong context.
 
AI isn't static. It evolves, and evolves a lot faster than human. From this thread.
index.php
If the goal is to wash that car, then driving it the mile to the car wash is usually the practical choice.

A 1-mile drive will add very little dirt compared with what the wash removes, and you'll avoid a 2-mile round-trip walk (plus however long you're at the car wash).

Walking might make sense if:

  • The weather is pleasant and you want the exercise.
  • The car is so dirty or dusty that you're trying to avoid adding any fresh road grime before washing.
  • There are parking or access issues at the car wash.
For most people, though, I'd just drive it there, wash it, and enjoy the clean car on the way home. ✨

It will be interesting to see how quickly end-users tire of this meandering bloviation as charging per token takes over.
 
One cool thing AI is doing is letting hobby projects progress rapidly.

The first PC-RPG I ever bought was Daggerfall back in 1996. As a DOS application that ran at 320x200 I kept waiting for a remake or remaster.

Daggerfall Unity was announced in 2014, I kept checking their news page for 5 years before it was in a state where I felt comfortable playing it, and even after 5 years of development it still wasn't 100% feature complete but very close. They didn't actually release the 1.0 version till 2023, so 9 years of development you had to wait on if you wanted to play 1.0

The latest trend in emulation is decompiling and getting raw code that way, then building native ports. Legend of Zelda Ocarina of Time has a full native PC port now thanks to this.

Silent Hill was released on the original PlayStation in 1999 and never had a PC Port. Siilent Hill has a decomp project and when it hit 99% decompiled a fan decided to code a port with Claude Code just for fun. Its progressed rapidly in only a couple of months including the author taking breaks here and there:

All major features work including rumble, music, in game movies. It is still rough around the edges and has crashes so emulators are a better experience for now but its only been in development for 4 months. At the rate its going I was say by the end of the summer at the latest it should be as good if not better then emulators.

If AI didn't exist this wouldn't be happening. There is another non-AI port that doesn't even have a single test release yet.
 
One cool thing AI is doing is letting hobby projects progress rapidly.

The first PC-RPG I ever bought was Daggerfall back in 1996. As a DOS application that ran at 320x200 I kept waiting for a remake or remaster.

Daggerfall Unity was announced in 2014, I kept checking their news page for 5 years before it was in a state where I felt comfortable playing it, and even after 5 years of development it still wasn't 100% feature complete but very close. They didn't actually release the 1.0 version till 2023, so 9 years of development you had to wait on if you wanted to play 1.0

The latest trend in emulation is decompiling and getting raw code that way, then building native ports. Legend of Zelda Ocarina of Time has a full native PC port now thanks to this.

Silent Hill was released on the original PlayStation in 1999 and never had a PC Port. Siilent Hill has a decomp project and when it hit 99% decompiled a fan decided to code a port with Claude Code just for fun. Its progressed rapidly in only a couple of months including the author taking breaks here and there:

All major features work including rumble, music, in game movies. It is still rough around the edges and has crashes so emulators are a better experience for now but its only been in development for 4 months. At the rate its going I was say by the end of the summer at the latest it should be as good if not better then emulators.

If AI didn't exist this wouldn't be happening. There is another non-AI port that doesn't even have a single test release yet.
I played my first serious game/space sim, Elite on my Commodore 64 back in 1985, and I eventually achieved Elite status, which for me was a huge accomplishment.
What I remember most was experiencing true 3D gameplay on a 2D screen for the first time. Combat required incredibly fast reactions and precise shooting. Docking your ship was also a real challenge you had to navigate through the rotating space station entrance in three dimensions and get the timing exactly right.

Imagination was key in this game. These days, games put you directly into the experience, but back then a large part of the adventure required lots of imagination.

Now, at 67, I don't really have the urge to play it again, but at the time it was an absolutely amazing game and a remarkable experience. It fitted easily on an floppy disk something like 40kb try that to day :facepalm:


 
Last edited:
One cool thing AI is doing is letting hobby projects progress rapidly.

The first PC-RPG I ever bought was Daggerfall back in 1996. As a DOS application that ran at 320x200 I kept waiting for a remake or remaster.

Daggerfall Unity was announced in 2014, I kept checking their news page for 5 years before it was in a state where I felt comfortable playing it, and even after 5 years of development it still wasn't 100% feature complete but very close. They didn't actually release the 1.0 version till 2023, so 9 years of development you had to wait on if you wanted to play 1.0

The latest trend in emulation is decompiling and getting raw code that way, then building native ports. Legend of Zelda Ocarina of Time has a full native PC port now thanks to this.

Silent Hill was released on the original PlayStation in 1999 and never had a PC Port. Siilent Hill has a decomp project and when it hit 99% decompiled a fan decided to code a port with Claude Code just for fun. Its progressed rapidly in only a couple of months including the author taking breaks here and there:

All major features work including rumble, music, in game movies. It is still rough around the edges and has crashes so emulators are a better experience for now but its only been in development for 4 months. At the rate its going I was say by the end of the summer at the latest it should be as good if not better then emulators.

If AI didn't exist this wouldn't be happening. There is another non-AI port that doesn't even have a single test release yet.
Unfortunately the workflow you describe makes the port a derivative work under copyright law, so vulnerable to a DMCA takedown or worse if the copyright owner is litigious. A safer route would be to have the AI write a specification based on the decompiled code, then separately create a port based on that specification, making it a clean room implementation of the function. That won't stop a litigious copyright holder taking action that'll cost you a lot, but it at least has a defence.
 
I played my first serious game/space sim, Elite on my Commodore 64 back in 1985

Elite is one of those games I kept meaning to try but never got into. When I got a PC in the mid 1990s I went back and tried some old 1980s games and I found most of them were not worth the setup time. I played a remake of the original Wolfenstein, NOT 3D lol. I don't even remember what it was called.

Elite Dangerous got excellent reviews but I never played it. If you like space stuff, FreeSpace 2 was excellent but sadly didn't sell well. It has a very dedicated modding community.
 
Unfortunately the workflow you describe makes the port a derivative work under copyright law, so vulnerable to a DMCA takedown or worse if the copyright owner is litigious. A safer route would be to have the AI write a specification based on the decompiled code, then separately create a port based on that specification, making it a clean room implementation of the function. That won't stop a litigious copyright holder taking action that'll cost you a lot, but it at least has a defence.

I don't know all the legalese of it but Nintendo is notoriously litigious, and they have yet to take down the Ocarina of time PC port.
 
Elite is one of those games I kept meaning to try but never got into. When I got a PC in the mid 1990s I went back and tried some old 1980s games and I found most of them were not worth the setup time. I played a remake of the original Wolfenstein, NOT 3D lol. I don't even remember what it was called.

Elite Dangerous got excellent reviews but I never played it. If you like space stuff, FreeSpace 2 was excellent but sadly didn't sell well. It has a very dedicated modding community.
For me, Elite was the first game that truly immersed me in a real-time 3D world, that fast/clever it was programmed on limited CPU speed. There was virtually no lag. The ship responded instantly to your controls. Must be programmed in pure maschine language to spare memory i guess. You had to think in three dimensions all the time basicly you became the game. Combat wasn't about spraying bullets everywhere; every burst had to be precise and effective to spare ammunition.

What I remember most is that as your rank increased, you became a bigger target. The galaxy seemed to take a personal interest in killing you. I must have died at least >50 over months before finally reaching Elite status.

So yes, Elite definitely tested your 3D skills to the limit especially surrounded by 15 hostile ships so how to turn how to roll in 3D effective/efficiënt etc etc. Looking back, it was probably one of the most demanding games I've ever played and all that in less memory than a modern screenshot :facepalm:
 
Last edited:
AI isn't static. It evolves, and evolves a lot faster than human. From this thread.
Despite efficiency gains, using better or more extensive models remains extremely costly. At the same time, we have already moved beyond the Pareto optimum when it comes to training.
Of course, there are niches where AI makes perfect sense. But for every conceivable application, it is not—at least for the moment—economically viable. Almost all AI companies are currently burning through cash, and economic metrics are being distorted by a massive circular economy of sorts. It reminds me of the VAT carousel fraud schemes in the EU.
When you factor in emissions (noise, exhaust fumes, water pollution, etc.) and the enormous consumption of resources (electricity, water), the overall picture looks, at best, mixed to me.
Another point worth making regarding AI development is that models are now being trained, to a significant extent, on the "slop" they themselves previously churned out. And since the traditional internet ("populated" by human-generated content) is shrinking in relative terms, the training base is unlikely to improve.

Don’t want this to come across the wrong way, I am absolutely not against AI. But statements like "it evolves" always read to me as if we were dealing with a genuinely evolving living being. What I think is absolutely brilliant is the performance of certain specialized, local AIs. Who would have imagined ten years ago what is now possible in photography thanks to AI-driven improvements in autofocus, noise reduction, upscaling, and so on?

YouTube recently added automated tagging of generative material, and revised requirements for content uploaders.
Unfortunately, this doesn't solve my main problem: AI slop music recommendations in the app. Consequently, I won't be renewing my YouTube Premium subscription (held since 2021) for my work laptop (all other streaming services are blocked ...) and my Spotify subscription (since 2011), and have switched to Deezer instead. There, you can report AI music, and you generally don't get AI music recommendations at all. It makes things a bit less convenient for me in the office, since I now have to stream to the desktop speakers via Bluetooth from my smartphone.
 
Despite efficiency gains, using better or more extensive models remains extremely costly. At the same time, we have already moved beyond the Pareto optimum when it comes to training.
Of course, there are niches where AI makes perfect sense. But for every conceivable application, it is not—at least for the moment—economically viable. Almost all AI companies are currently burning through cash, and economic metrics are being distorted by a massive circular economy of sorts. It reminds me of the VAT carousel fraud schemes in the EU.
When you factor in emissions (noise, exhaust fumes, water pollution, etc.) and the enormous consumption of resources (electricity, water), the overall picture looks, at best, mixed to me.
Another point worth making regarding AI development is that models are now being trained, to a significant extent, on the "slop" they themselves previously churned out. And since the traditional internet ("populated" by human-generated content) is shrinking in relative terms, the training base is unlikely to improve.

Don’t want this to come across the wrong way, I am absolutely not against AI. But statements like "it evolves" always read to me as if we were dealing with a genuinely evolving living being. What I think is absolutely brilliant is the performance of certain specialized, local AIs. Who would have imagined ten years ago what is now possible in photography thanks to AI-driven improvements in autofocus, noise reduction, upscaling, and so on?


Unfortunately, this doesn't solve my main problem: AI slop music recommendations in the app. Consequently, I won't be renewing my YouTube Premium subscription (held since 2021) for my work laptop (all other streaming services are blocked ...) and my Spotify subscription (since 2011), and have switched to Deezer instead. There, you can report AI music, and you generally don't get AI music recommendations at all. It makes things a bit less convenient for me in the office, since I now have to stream to the desktop speakers via Bluetooth from my smartphone.
The funny thing is that being a "living" organism doesn't automatically mean continuous improvement either.

Quite some people I've met seem to become more rigid an not only because of age. Memory declines, frustration increases, people stop paying attention to what is actually being said, fall back on egocentric reasoning, or generalize because they lack intelligence (or absent of Abstract thinking) or willingness to engage with counterarguments and facts.

In that sense, "being alive" is not necessarily a guarantee of meaningful evolution. Sometimes it's the opposite.

When I notice that, I usually don't argue for long. Most of us probably recognize it within seconds minutes. It is what it is.:facepalm:

Adaptation and cognitive habits decline over time due to entropy (wear) I suppose that already started from birth also for me.

So I don't see the word "evolves" as something that should be reserved exclusively for living beings lot of times it decreases. Systems can evolve too, even if the mechanisms are completely different.
 
Last edited:
  • Like
Reactions: NTK
Don’t want this to come across the wrong way, I am absolutely not against AI. But statements like "it evolves" always read to me as if we were dealing with a genuinely evolving living being. What I think is absolutely brilliant is the performance of certain specialized, local AIs. Who would have imagined ten years ago what is now possible in photography thanks to AI-driven improvements in autofocus, noise reduction, upscaling, and so on?
When I used the word "evolve", I am more referring to AI as a technology, and the word is used regularly in this context. But let's face it, we have a very significant (if not majority) chunk of the best minds of this generation working on it.
 
I mentioned it a few posts ago, as of today you can complete Silent Hill 1 on PC in all x86 code. No emulators.

Its still very rough around the edges, visual glitches, sound timing issues, but you CAN finish the game. 4 months of work from one programmer + AI and a lot of logs generated from testers.

 
Back
Top Bottom