AI and it's effect on your music, your job and the future

[p0ke]

re: AI taking over the roles of engineers, just imagine how much worse this comic would be with current AI interpreting client demands. I see some MC Escher-type designs getting made and production just hanging up trying to interpret it

.View attachment 147941

This is exactly what I was talking about earlier. Yeah, sure, the issue could also be that the company just doesn't understand the customer, but in my case most times the customer just doesn't know what they want.

 

[SalsaWood]

Isnt AI destroying the planet faster with its required resources to run…?

The amount of data currently stored and transferred with zero other-than-cursory value, and the amount of pedantic computation going on for the simple sake of computation, is astronomical. We're talking millions of lifetimes worth of brain rot, primordial, or single use content and data operations. AI are like a drop in the ocean so far.

 

[TedEH]

AI are like a drop in the ocean so far.

Are they? Storage doesn't burn electricity in the same way. An astronomical amount of useless data floating around in data centers causes a certain amount of passive cost just to maintain the centers it lives in, but AI, crypto, etc., actively burn a whole lot of power on top of the same passive center costs. On a smaller scale, it would be like comparing having a media server with a bunch of harddrives that mostly sit idle, vs. running video benchmarks on a 3090 indefinitely.

There are some articles out there that estimate the power cost for AI, but what's interesting is that they're just measuring the computation cost on the hardware, not accounting for the storage or the rest of the attached hardware that goes with it. This article from the Verge claims nvidia GPUs used for AI cost about .5% of global electricity use, compared to 1-2% used for "everything that is the internet". https://www.theverge.com/24066646/ai-electricity-energy-watts-generative-consumption That doesn't paint a good picture, even if it's just a guess - far from "a drop in the ocean". That's saying AI has cost somewhere from a quarter to half as much as the whole internet.

 

[Demiurge]

I'm sure that in the future AI will help us fix all of the problems that creating & implementing AI has caused. Don't worry- it's the future's problem. In the meantime, please enjoy this picture of a toddler building a sand sculpture of a soldier kneeling before a cross with caption "why dont pictures like this trend?"

 

[narad]

It's a bit of a negotiation. AI training costs are high and will continue to be high for the foreseeable future until there are other ways of training or improving them (evolutionary algorithms, possibly). But what do you get in exchange for that compute cost? If AI helps make a breakthrough cancer treatment or design a more efficient battery, you get these gains and you get them in perpetuity. And what do you get for other such expenditures of electricity?

 

[TedEH]

But what do you get in exchange for that compute cost?

An industry of attention-bait slop that preys on the unaware boomers of facebook for ad shares?

And what do you get for other such expenditures of electricity?

Light? Heating? The Netflix part of Netflix and chill?

I kid - I get that AI has the capacity to do some real good, but that doesn't negate its cost. Or the slop that comes with it.

 

[Moongrum]

If AI helps make a breakthrough cancer treatment or design a more efficient battery

I agree with your sentiment, but I doubt the bulk of electricity usage for AI as a whole is for the betterment of our lives, but instead whatever bullshit product microsoft/google/etc is trying to peddle to consumers and businesses for profit rather than some noble cause (although the two can intersect).

I wonder if your optimism and our hesitancy comes from you being a researcher, being on the cutting edge, and believing in the work you do. Whereas those of us who are programmers/software engineers/whatever, work on products we don't care about or believe in. Maybe it's only me, though lol.

I work for a giant corpa for their business travel division, and we were asked at the beginning of the year to figure out how to incorporate AI into what we are doing, and upper level management always talks about how this is such an exciting time for our company and how we're going to revolutionize the industry with AI, blahblahblahblah while I have no idea what we are actually going to do with it. I asked my manager what it is our company is doing with AI, and he said I just focus only on the work that's assigned to our team. Perhaps you can see where my fatigue of it all comes from

 

[narad]

I agree with your sentiment, but I doubt the bulk of electricity usage for AI as a whole is for the betterment of our lives, but instead whatever bullshit product microsoft/google/etc is trying to peddle to consumers and businesses for profit rather than some noble cause (although the two can intersect).

Nah, but it's all on the path to greater understanding. Regardless of who is doing it, we need people to train these models to better understand the models and the training process, and in the context of big-tech, the competition seems like a good thing.

I wonder if your optimism and our hesitancy comes from you being a researcher, being on the cutting edge, and believing in the work you do. Whereas those of us who are programmers/software engineers/whatever, work on products we don't care about or believe in. Maybe it's only me, though lol.

Well it's not like I've lost sight that a lot of the human-facing AI stuff is going to be a bunch of business buzzwords and hype, thrown around with little real understanding of how the technology can be applied, both now and in the future. I still get a good chunk of that myself. But it's been very weird to me as a researcher to see the sentiment emerge around AI being fairly negative, when from my pov it's probably the most important human technology to ever be created, and crucial for continued scientific development. The importance of scaling up the net productivity of human civilization far outweighs the internet slop or electricity costs. That's how we're going to solve the big problems facing society now, in healthcare, in the environment, in logistics, in space travel, etc. I understand that the big picture may not be at the forefront of everyone's mind every time they encounter some AI-generated annoyance, but it still surprises me how rare it is to even have it considered.

 

[SalsaWood]

AI, crypto, etc., actively burn a whole lot of power on top of the same passive center costs.

I was actually considering crypto mining as computation for the sake of computation compared to strictly AI. AI will probably push lots of forms of masturbatory computations further, though. AI has come to basically mean software, which we now call apps, and used to be called programs.

 

[nightsprinter]

I'm sick of AI stealing my riffs

 

[gabito]

I remember discussions similar to this one but about the internet in the late 90s, early 00s.

And here we are, using the thing a lot of people were afraid of 25 or 30 years ago without even thinking about it.

In a few years nobody will know AI is there and everybody will be using it without even noticing.

Personally? As a developer it helps. It's like a supercharged search engine, manual, companion or whatever. I don't see it replacing ALL THE JOBS, because it helps but we're immensely stupid and complicated to a degree no computer will probably ever understand It might help me with scripting, programming, configuring things, etc., but there's a lot more to development, and AIs (still) can't solve a lot of stuff we have to deal with daily.

I understand the concerns about all the power these things consume, but so does Netflix, Google, Amazon and a lot of other stuff we all use daily. The world is ruled by corporations, they don't give a fuck, and I'm not sure we can do something about it. Probably not.

Anyway, if Idiocracy turned out to be a documentary maybe The Matrix will turn out to be one too eventually. If that means I could learn kung fu or how to play like Paul Gilbert just by downloading something into my brain, so be it.

 

[TedEH]

we need people to train these models

The word "need" is carrying a lot of weight here. There's no need for most of the AI out there. ChatGPT is not a need. Smart appliances are not a need. Really advanced minecraft bots are not a need. 400 bazillion propaganda photos with too many fingers are not a need. Medical and science advancements could be had using machine learning without abstracting it out to chat bots and generalized job automation and trying to humanize computers.

AI has come to basically mean software, which we now call apps, and used to be called programs.

No it hasn't. Not generally enough to put it that way. There are some folks trying to make this happen, or lying about how much "AI" is in their software, and they're being called out for it.

 

[BlackMastodon]

And what do you get for other such expenditures of electricity?

Curious about the number of Hitachi's duct taped to folding chairs that could be powered instead.

 

[BlackMastodon]

But it's been very weird to me as a researcher to see the sentiment emerge around AI being fairly negative, when from my pov it's probably the most important human technology to ever be created, and crucial for continued scientific development. The importance of scaling up the net productivity of human civilization far outweighs the internet slop or electricity costs. That's how we're going to solve the big problems facing society now, in healthcare, in the environment, in logistics, in space travel, etc. I understand that the big picture may not be at the forefront of everyone's mind every time they encounter some AI-generated annoyance, but it still surprises me how rare it is to even have it considered.

I wish I had your optimism, but I feel like this is the reality:

The world is ruled by corporations, they don't give a fuck, and I'm not sure we can do something about it. Probably not.

I feel like any hope of AI solving humanity's roadblocks and problems is like the last act of Don't Look Up. Corporations and lawmakers and whoever else is set to profit the most from AI will absolutely knee-cap any chance for progress just to make a buck. I really don't expect it to make anything better, only make the small percentage of rich people richer. In the meantime, more job cuts in the tech sector when the cost of living is higher than it's ever been and people are already struggling to get by with only 1 job, all because these shitdicks see dollar signs instead of human beings. This isn't explicitly AI/machine learning's problem, just a human problem that is going to keep on rollin'.

 

[crushingpetal]

This highlights some issues nicely: NPR "10 Reasons why AI may be overrated":

https://www.npr.org/sections/planet-money/2024/08/06/g-s1-15245/10-reasons-why-ai-may-be-overrated-artificial-intelligence

 

[crushingpetal]

And... the link above doesn't even mention the issue of "model collapse" .

My current view (subject to change at will): current generative AI will be "disruptive" and it is mediocre. The future is a sea of cheap garbage.

 

[narad]

The word "need" is carrying a lot of weight here. There's no need for most of the AI out there. ChatGPT is not a need. Smart appliances are not a need. Really advanced minecraft bots are not a need. 400 bazillion propaganda photos with too many fingers are not a need. Medical and science advancements could be had using machine learning without abstracting it out to chat bots and generalized job automation and trying to humanize computers.

You don't need to abstract it out to chatbots for it to be AI. Advancing medicine and science with machine learning -is- AI for all intents and purposes. People need to stop equating ChatGPT to all of AI. We need the ability to read and aggregate scientific studies on a superhuman scale to make reasonably fast strides in advancing our technology -- the literature is constantly growing, and no person can really be up-to-date in all but the most niche of scientific topics. AI is the only currently plausible solution to that problem, and that is why I consider it a need.

And... the link above doesn't even mention the issue of "model collapse" .

Model collapse is actually not that much of a problem. The difference between a model trained on a 2023 internet snapshot and a 2025 snapshot isn't going to be breakthrough super intelligence in that extra 5% or whatever training data, just familiarity with 2024-2025 concepts. I think we have basically all the text-of-the-internet data we're ever going to need when it comes from gleaning an understanding of the world from raw text.

 

[TedEH]

People need to stop equating ChatGPT to all of AI.

I can agree with that.

I'm of the opinion that a lot of the things we're trying to solve with AI could also be solved without AI. AI is the hammer du jour, so every problem looks like a nail right now.