it would mean that theory (which is not hindered by batteries) would be always vastly outpacing technology, which it isn't.
The AI debate may become moot sooner than we think
- Thread starter tanstaafl
- Start date
- Apr 30, 2017
- 4,553
- 20,272
this whole meta-discussion about employment and development pipelines reminds me of the 60s when there was a push for proving programs through formal logic, thinking we can automatize producing flawless code with very similar repercussions of making humans obsolete. obviously it turned out not to work and now that whole thing is just a footnote in the history of computing.
and that was a MUCH stronger idea, AI doesn't even attempt to produce flawless code and has no mechanisms for avoiding that. we're now more just throwing shit on a wall and hoping more than 40% will stick.
and that was a MUCH stronger idea, AI doesn't even attempt to produce flawless code and has no mechanisms for avoiding that. we're now more just throwing shit on a wall and hoping more than 40% will stick.
- Jun 19, 2019
- 8,454
- 32,296
This whole meta-discussion about employment and development pipelines reminds me of the 1860s when there was a push for tractors to replace farm workers.
You must be registered to see the links
To be fair he isn't wrong.
What people are missing from the puzzle is that AIs are currently only trained to "predict" pixels or word tokens.
Once AIs are getting trained to "predict" vertex and triangles through model topology data, 3D generation would be solved pretty much instantly.
The greatest pain of models like Daz is with morphing the characters to look distinct enough from other games, this is why you only see a few characters in games like Daz where you can have hundreds of characters in games based on Koikatsu/Honey Select.
With AI you can have morphing based on prompts or reference images basically instantly.
And you only need to generate a 3D character once, so there is no issue of consistency.
And if you have 3D generation you can feed that back into the 2D image AIs to fix whatever it has with character consistency things like multiple fingers and parts spawning in the wrong place.
What people are missing from the puzzle is that AIs are currently only trained to "predict" pixels or word tokens.
Once AIs are getting trained to "predict" vertex and triangles through model topology data, 3D generation would be solved pretty much instantly.
The greatest pain of models like Daz is with morphing the characters to look distinct enough from other games, this is why you only see a few characters in games like Daz where you can have hundreds of characters in games based on Koikatsu/Honey Select.
With AI you can have morphing based on prompts or reference images basically instantly.
And you only need to generate a 3D character once, so there is no issue of consistency.
And if you have 3D generation you can feed that back into the 2D image AIs to fix whatever it has with character consistency things like multiple fingers and parts spawning in the wrong place.
- Apr 13, 2021
- 1,832
- 1,500
The human brain is easy fooled.
If you think about it, anything not in your view, the brain makes up.
And so, even if you see AI generated content, you sometimes miss the unreal things in an image.
AI will be soon so good that you can't differentiate between real or fake.
Now, some will say this is great because i can fuck around with images and make my personal porn actress.
Others will be reserved about it.
No matter where you stand, even you can be made into a fake AI generate image.
Perhaps it can be liberating later on when everyone has the same tools available to them to create fakes of others.
But of course, this won't happen because bad people usually have the upper hand.
I think some version of fakes showed that AI has had problems with details. Like fingers or toes. Overall the images looked good but the details where the give away.
Zombie teeth or fingers.
But the evolution of AI isn't done. Eventually it will be so good to present a real fake image where you can't tell the difference. Perhaps you know how the real person looks like but eventually the fake personality will be omnipresent.
On a technical level, i am sure it will be perfection. On a moral and personal level, it will be scary.
It will be like, i am artist because if can operate an AI. Even though you are not an artist. So nerds will be able to claim they are.
In some sense i am glad i will die one day as i am not sure if i like that future at all.
- Oct 29, 2018
- 1,162
- 1,633
I'm pretty sure this thing I watched about batteries was in the 00s or 10s, not the 20s. Because, I can't remember much about it at all other than it's main point was that if battery technology took a leap forward so would the technology we have.
Edit: Maybe a non sequitur, but I also remember it making the point that the limitations of current battery technology is why we don't have earbuds with the processing power of an iphone. etc. That kind of thing.
Last edited:
more cold water onto the AI hype. among other things the video goes quickly through cases where google, tesla, GM, openai and many other big companies have been caught fabricating their results and often even using humans to fake AI behavior.
You must be registered to see the links
Who is the audience for this video? People that actually think ChatGPT is sentient?
There is certainly massive overblown hype, but the underlying tech is real. And it's not just about chatbots that horny kids can talk to. The unifying technology underneath all the various implementations of "AI," be it art, chat, self-driving, or what have you, is deep-learning and machine learning and the ability to shift through massive amounts of data and use that as a basis to generate new outputs. The fact that all major tech giants are investing heavily into this means that they are all banking on it increasing productivity and their bottom line.
we already went over how there's a study showing that the diminishing returns of increasing teaching data and models is unlikely to be enough to reach that wide generalization people are hyping.
what huge corporations throw money at is largely decided by seeking monopoly (by buying out competition then shutting production) and whatever some charlatan consultant without technical expertise told the suits at the latest corporate leadership seminar/junket getaway. the people who call the shots have NO idea how the tech works. they just buy into hype just in case it works.
but from the looks of it it's just a question of time before openai implodes. and when it goes I don't know what's left of the AI boom.
- Oct 29, 2018
- 1,162
- 1,633
You can stand on your soapbox as long as you want, it'll be a good view for you to watch as AI continues on despite what you are trying to convey.
I've never said AI wouldn't continue, but I've repeated ad nauseum that all points to LLMs not being able to take us to the 'next level'. I even said in this thread: "the good news is such strategies are possible as our brain already do it effortlessly, but the problem is we have no idea how."
- Oct 29, 2018
- 1,162
- 1,633
You don't even know what the next level is. Neither do I. Is it solidifying core concepts? Is it expansion into higher mathematics more accurately? Is it a break through in efficiency and storage affecting the memory of AI going forward? Is it the jump into 3d as talked about earlier in this thread? You're creating your own echo chamber by looking up every possible negative and not paying any attention at all to the directions AI is going.
You cover your own ass by saying "Well it may do this but all this over here says it won't." Pick a lane.
Edit: Also, most of the "falsifying" you're talking about are actual AI results that don't show the actual prompting used. Like one where it was using visual stimuli and getting responses from the AI reading text and monitors. That was actual results but with several images left out of the gathering. A little misleading, but there was no actual faking of results. I'll see if I can find that example.
Last edited:
Does it require a generalized zero-shot model that can do anything for the AI boom to take off?
This is what pisses me off about the Anti-AI muppets.
How many layers does a artist use nowadays for a single art piece?
How many reference images do they use?
You think you can draw a whale without using "internet images" when you haven't seen a real whale in your life? You think you can draw that right now from memory without getting the details wrong?
The Revolution already happened, now is only a question of finding the right Process and Workflow in how we use them.
We don't fucking need AGI for that. We are Humans that are enterprising and we will figure something out.