This is fantastic. I've been playing with Confyui for a while and I was wondering if MLOPs could be a similar tool, but within the Houdini interface - which would be just perfect! The problem with Confyui, at least for me, is the amount of information you have to put in the node itself - since it doesn't have a "Parameter view" like Houdini - and things can get visually confusing like this. I can't wait to start playing with MLOPs, thank you so much for this Moritz! - Alvaro
Learning a lot of this from scratch and one thing I struggle to establish is what kind of math is required to do x y or z (learning everything would take years), so 18:00 thank you very much guys, linear algebra is another piece of the puzzle.
Good discussion! One comment: somewhat counterintuitively, i think the rise of A.I. could take things back in a more "analog" direction. If you take a Pixar movie, and you watch it from start to finish, it's perfect, you won't (without pixel-peeping) find a single glitch or defect. Why i this? Because the people at Pixar have decided this must be so. But looking back at animation before digital, it wasn't perfect, there were plenty of rough edges, yet people seemed to enjoy it equally. So could it be that a new generation could learn to accept this less-than-perfect quality? Because what they get in return is something much more personally tailored for them, you could almost call it "story-on-demand". Perhaps you're sick of the formulaic Hollywood story structures, and you want personalized, rendered animation movies with stranger stories, well, that'll be possible. But it won't have that Pixar perfection, that's why i call it more "analog". Perhaps people will prefer that to the one-size-fits-all perfection? In the same way, computer programs may become more analog. Like a human doing a task won't be 100% repeatable or predictable each time, perhaps the new generation of A.I. applications will also have rougher edges, but will be more useful, because they'll be tailor made to our needs. So i want to push back a little on the general vibe of this pod that use of A.I. has a ceiling because it won't get you all the way to that perfection that is the current norm. Because it's possible that people will no longer demand that perfection. I think this is the danger of the current way of thinking about A.I. with regard to job replacement. Just a thought!
Firstly the word AI is a marketing misleading, it’s actually mostly linear algebra. Secondly, you say « analog », but real « analog » systems are the real randomness. The kind of randomness you can’t really get in a computer, but just try to simulate it to the best, so it seems random enough. Try to use stable diffusion with the same seed and the same prompt and parameters and you will find, that actually you can get the same image every time. So in fact it’s not more random than we program or accept it to be. From my perspective It is a new tool, which can do good or bad results, yet quite difficult to control when having a very specific demand.
Amazing project. something that was not mentioned about why someone would want to use stable diffusion in houdini: being able to create variations and automating tasks with any level of flexibility with pdg. Still didnt dive in the discord, very curious to see what people is doing with this tool
So, so good! I will make sure to listen in on all you have to say! pure gold! im also blessed having a gf that is a doctor in algebra so i get that for free as well :D
Didn't listen whole show, might mentioned, but If you know how much $$$$$$ need to train chatGPT etc you will be very surprised.... AI should be a helper, not a "replacement". Thinking of that AI will write programs for you, automatically devalue your skills and knowledge and will directly affects your paycheck. So, not much "fun"
well for basic things possibly, however it's no different to auto correct, most people don't know how to spell now, but jobs where its required such as lawyer for example if its not exactly correct, much like code or a simulation your client won't be happy with the end result.
Bro, have you personally used any AI yourself? Because if you had you'd know AI simply can't touch human expertise, not only does it straight up lie, but it also writes bad code. If you used it and didn't know anything about vex for example, you'd never deliver anything of use
@@TRUEbASNER many people, general public, think about AI as a magic bullet. I'm working professionaly for machine learning stuff, not using but training. Not regarding to the code, just look at MidJourney - it's devalue and killing all concept art as a job, thousands of people layed off or will be by that.
Then we can actually use these tools to figure out the real problems. Fusion, space travel, curing cancer. It would be a real shame that you can't buy 5 bags of chips instead of 4. Jesus...
This is fantastic.
I've been playing with Confyui for a while and I was wondering if MLOPs could be a similar tool, but within the Houdini interface - which would be just perfect!
The problem with Confyui, at least for me, is the amount of information you have to put in the node itself - since it doesn't have a "Parameter view" like Houdini - and things can get visually confusing like this.
I can't wait to start playing with MLOPs, thank you so much for this Moritz!
-
Alvaro
Learning a lot of this from scratch and one thing I struggle to establish is what kind of math is required to do x y or z (learning everything would take years), so 18:00 thank you very much guys, linear algebra is another piece of the puzzle.
I appreciate y'all for this! Excited for more nerd rants to come
Still trying to wrap my head around the usefulness of MLOPs, I have it, but I've never felt a need to use it beyond the tutorials.
Amazing Thank you guys, this is very insightful!
Good discussion! One comment: somewhat counterintuitively, i think the rise of A.I. could take things back in a more "analog" direction. If you take a Pixar movie, and you watch it from start to finish, it's perfect, you won't (without pixel-peeping) find a single glitch or defect. Why i this? Because the people at Pixar have decided this must be so. But looking back at animation before digital, it wasn't perfect, there were plenty of rough edges, yet people seemed to enjoy it equally. So could it be that a new generation could learn to accept this less-than-perfect quality? Because what they get in return is something much more personally tailored for them, you could almost call it "story-on-demand". Perhaps you're sick of the formulaic Hollywood story structures, and you want personalized, rendered animation movies with stranger stories, well, that'll be possible. But it won't have that Pixar perfection, that's why i call it more "analog". Perhaps people will prefer that to the one-size-fits-all perfection? In the same way, computer programs may become more analog. Like a human doing a task won't be 100% repeatable or predictable each time, perhaps the new generation of A.I. applications will also have rougher edges, but will be more useful, because they'll be tailor made to our needs. So i want to push back a little on the general vibe of this pod that use of A.I. has a ceiling because it won't get you all the way to that perfection that is the current norm. Because it's possible that people will no longer demand that perfection. I think this is the danger of the current way of thinking about A.I. with regard to job replacement. Just a thought!
Firstly the word AI is a marketing misleading, it’s actually mostly linear algebra.
Secondly, you say « analog », but real « analog » systems are the real randomness. The kind of randomness you can’t really get in a computer, but just try to simulate it to the best, so it seems random enough.
Try to use stable diffusion with the same seed and the same prompt and parameters and you will find, that actually you can get the same image every time. So in fact it’s not more random than we program or accept it to be.
From my perspective It is a new tool, which can do good or bad results, yet quite difficult to control when having a very specific demand.
oh and the Terminator eyes just popped up after lolz!
are you guys ever going to make tutorials or nerd rants again? or will it be the chris kopic channel for all eternity now?
Amazing project. something that was not mentioned about why someone would want to use stable diffusion in houdini: being able to create variations and automating tasks with any level of flexibility with pdg. Still didnt dive in the discord, very curious to see what people is doing with this tool
Is there anything like image-to-material maps coming? That's the only reason left to buy Sampler now.
thanks for the nerdfo though =)
What is that ML tutorial series that was mentioned? at 19:12
3blue1brown youtube series on linear algebra and one on neural networks as well. Incredibly useful stuff!
So, so good! I will make sure to listen in on all you have to say! pure gold!
im also blessed having a gf that is a doctor in algebra so i get that for free as well :D
Make sure to keep her around! :)
Didn't listen whole show, might mentioned, but If you know how much $$$$$$ need to train chatGPT etc you will be very surprised.... AI should be a helper, not a "replacement". Thinking of that AI will write programs for you, automatically devalue your skills and knowledge and will directly affects your paycheck. So, not much "fun"
well for basic things possibly, however it's no different to auto correct, most people don't know how to spell now, but jobs where its required such as lawyer for example if its not exactly correct, much like code or a simulation your client won't be happy with the end result.
Bro, have you personally used any AI yourself? Because if you had you'd know AI simply can't touch human expertise, not only does it straight up lie, but it also writes bad code. If you used it and didn't know anything about vex for example, you'd never deliver anything of use
@@TRUEbASNER many people, general public, think about AI as a magic bullet. I'm working professionaly for machine learning stuff, not using but training. Not regarding to the code, just look at MidJourney - it's devalue and killing all concept art as a job, thousands of people layed off or will be by that.
@@rendermanpro OK, so they'll have to adapt and do something else instead. That's how it goes.
Then we can actually use these tools to figure out the real problems. Fusion, space travel, curing cancer. It would be a real shame that you can't buy 5 bags of chips instead of 4. Jesus...
your audio is out of sync!
it's not. it was some weird glitch. sorry.