each week, I spend more and more time with it. I think the charts thing was so smart of them because you can see meme-y things like the recipe songs pop up and try them on your own
Subscribed. I've been using computers since 1972, my first program was a basic demonstration of AI of a common program called "Animals", and I have been aware of neural network AI since the mid 1980s. Generative AI has taken a huge leap, but I don't know how some people that know how they work think that they are any kind of sentient, or ever will be.
Welcome! Yeah, it's a complicated question and I guess the biggest thing will be what happens at larger and larger scales. I think so much of the question of sentience is in part because we don't really know how we're sentient either?
I have a model 3 with FSD v12 and it is able to take me around my area with no interventions probably 90% of the time now. V12 does some crazy good maneuvers I never would’ve expected in the previous 2 years of testing FSD beta
I'm rate limited on GPT-4 and Opus most of the time since I have them pumping out so much synthetic data for model distillation and fine-tuning. The people who say it doesn't work are probably just inexperienced at prompting. There are some really neat insights gained though, like when distilling a model on synthetic text data from GPT-4, you get better results if you sequence the training to first use fake children's literature before phasing into more complex topics. Like teaching a student, it helps to arrange the content pedagogically. For synthetic data generation, it seems the key is to present a precise outline for the larger content structure and instructions for how it should express writing objectives before commencing on any piece. Without a human in the loop, the danger isn't so much that it devolves into inaccuracy, but rather it will lose sight of hierarchical structure and continuity of purpose, becoming redundant or overextending beyond a topical scope's definition. It's also helpful to embed notations that contextualize information metacognitively, using formal nomenclatures like Bloom's taxonomy from established curriculum writing, so that any content is presented mindfully as a learning experience. Inaccuracies aren't really a problem for anyone who knows how to employ tree-of-thoughts and if implementing breakpoints for iterative critical review and revision subsequences. The greater context length on Opus makes a lot of things possible now that weren't just a few weeks ago, but I still prefer GPT-4 (using a committee of experts) for most of the preparation tasks. For synthetic video, I'd guess a big focus is on procedurally generating cinematics in Unreal 5 and training on those.
@@AIForHumansShow Here is a link to a starter template I made for a graduate level technical textbook drafting. Simple xml markup seems fine for interpretation. This should be relatively adaptable for other purposes, and once objectives are well defined, you can embed notations in the content for referencing them directly to improve revisions. Some links are included for my GPT review committee, but making those for yourself is one of the most interesting parts. Lowkey, Mythos is my favorite. I understand also that OpenAI has their own rubric and embedding notation for training student models, but I'm not caught up on the specifics. My larger aims later are to present knowledge domains or simulated spaces as networked graphs of text elements with multiple agents collaboratively navigating it for continuous section-wise iterative review and revision. For the simulated spaces, it would functionally be like a game of Zork where the NPCs build the world themselves from the inside out, but the text-based spatial descriptions can be formatted into image generation prompts, and later too adapted into procedural level generation for Unreal. I'm still brainstorming though! And really busy with school. Template link: www.dropbox.com/scl/fi/x93vq1y48p5lj9ezqscgy/educational-content-generation-template.docx?rlkey=tu6ne7yn4hblrqi8owomh4ppd&dl=0
I've never heard the John Henry story with him being a coal miner. I've always only ever heard it with him pounding railroad spikes. Very interesting! BTW I was lucky enough to catch the last episode which was the first show? I think lol. Anyways, you know what's funny is I recognized Kevin immediately from when I was a kid. You've aged well Kev and your just as funny as I remember and more. Definitely subscribed! Great work fellas!
Lol hey now I misremeber things all the time. Also regardless of what job he was doing, you got the moral of the story accross. But since your a celebrity were gonna chalk this up to a simple "mandella effect" 😅
@@DubStepKid801 Hahahaha me as 'celebrity' is as off as coal mining vs railroad spikes but I appreciate it. truthfully, I should've remembered it's one of my favorite American Myth stories.
The idea that you can choose specific areas to edit in AI is pretty awesome. But what really got me was the concept when Purzbeats mentioned we should ask AI if it wants to work with us...
Yeah it sure is fun to play around with - we missed it for this week because we record on Tuesdays but will def cover next week. Oooooh maybe EurovisAIon Song Contest?!?? Suno vs Udio??
@@AIForHumansShow haha that would be epic! Actually a song contest is a great idea, see who can come up with the wackiest most radical baja blastin tune!
True Story incoming (this is Gavin)... I love my Mustard hat so much that I nearly wore it out (the first one, this is the second one I have). It was a gift from my wife who knows how I love weird absurd dumb things to wear. My favorite shirt of all time was a white t-shirt that just said "THE MERGE" in bold red font. You get the idea. My daughter thought she'd see what other hats were out there and lo and behold there WAS a red hat that says 'ketchup' in white lettering that normally would look very cool and fun and I thought WOW what a gift I love this. I wore it out and got a ton of weird looks, include a few people that literally looked at me with what I can only describe as disgust. I didn't understand at all until I saw myself in the mirror briefly and realized it looked a LOT like another very famous red hat with white lettering. Needless to say, I did not wear that Ketchup hat again but it still sits on my shelf beckoning me.
Just found out about your show. Thoroughly informed and well entertained. You have my subscription sirs. Also, i kinda find it ironic that uber and others tried the robo taxi path and it might end up being taken by Tesla who solved rhe hardest part of the problem and figured the other side is easier so they could just add it on. I can see a whole bunch of orher scenarios where that plays out similarly. It'll just be held back by internal bureaucracy if others dont
First, thank you for watching and subscribing! Second, yeah it's so funny to think that it was just going to be ubiquitous (and maybe ultimately it will be!) but I certainly hope we can hit it within the next five years.
Did you hear about the controversy about AI art being used in the movie Late Night With The Devil? I only heard a little about it but supposedly people are calling for a boycott of the movie in protest. I was hoping to get your thoughts.
We did hear about this - we had it on our rundown a few weeks ago but decided not to talk about it mostly because we had a few other things pop up. Tbh, it felt overblown in the way that many of these AI stories are. It wasn’t even a lot, like a few particular images. There’s just a lot of anger amongst certain creative crowds around AI in general and we totally get it. But as we discuss in this weeks show, ultimately it’s a tool that can be used by human creativity to great effect.
@@gavinpurcell now are you just saying that? Or have you actually watched my terrible content lol. I'm just happy my continued engagement has been help y'all roll the boulder up the algorithmic mountain. You guys doubled from 3k in like what, a month?!
You need to add your own names in the video description. C'mon, personal branding and SEO!!! AI not gonna do that for ya. Watched you, and loved you, but I have no idea how to tell other people who you are, as there are no presenter names in your description.
we kindly disagree. an artist is someone who works hard at their craft and feels something strongly about the work the put out into the world. ESPECIALLY what Purz is doing is that and isn't just machine in and machine out. Go watch one of this how-to videos and tell us what he's doing isn't specifically art technique that's something MUCH different than riding in a plane
@15m he's totality 100% right., but just scraping data isn't the problem imo., you are no more than a bunch of haphazardly thrown together algo sets yourselves, you scrape as you go cuz you can't absorb everything, AI can only record and react though, cuz no feelzbag hardware detected.... hooman gooputers lack lots of things., AI simply allows for those things without implanting chips & wifi in your brain.. but we need clones and a place to chill while we wait for them anyway,. AI is what will facilitate that too. soo FtheDs imo. i want immortality by 2042., get onn it
@@AIForHumansShow full on fresh meatsacks imo., We should all have a 25yr old clone for parts already ... thats the easy part,. so i see no reason why you couldn't do both and i would think both would become possible at the same time., cuz its going to be a convergence of all our tech like a singularity that blooms in to "stuff".., data storage ,neuro links, quantum computing, AI*, Crisper and the like,, all of it is coming to a head as we speak and soon 💥 all of the above and way more becomes possible.,. In fact what's on that list should be thoroughly explored immediately. we need to min/max this task & Elon being in charge of neurolink horrifies me..
@@gavinpurcell seems like it ,. but elon isn't even pushing for it., he's doing some other thing.. this is an A+B+7+*+j = immortal situation., only a few even know the thing they're working on might = immortality at the end of the day in 20 years,. everything from crypto to crisper might have a part to play
Another great show, I am always learning something and laughing every episode. Real quick...anybody else think that the 1897 Gambit photo looked a little like Kevin? Kevin how old are you...really?
@@gavinpurcell 🤣🤣 Maybe, just, 🤏...lol...(This was the very first video I watched this morning. As I was making coffee. Before the coffee kicked in. When everything else in the world is moving just a little too fast!)
@@gavinpurcell It was literally me, being an old man, waiting for my coffee, thinking, "These darn whipper snapping kids and their darn energy! Back in MY day...." 🤣🤣🤣
Some really interesting research is coming out about using AI to review its own responses and then use that review to make better ones. Its called Agent prompting and researchers were able to make GPT 3.5 get results on par with GPT 4, albeit taking much longer per response. This can probably be used as a loop to create synthetic data. Use agents of your current models to create better responses, train new model on those responses, then use agents of your new model to create even better responses and repeat.
Thanks for having me, I had a blast!
thanks for joining Purz! def gonna dive more into comfy today
Def! Everyone should check out Purz's tutorials on his YT page while you're here as well!!
subbed!
Thanks so much! I've needed to find a starting place and will use this as guidance and inspiration.
incrdible interview. Purz is refreshingly articulate and grounded talking about this whirlwindy magic pixie dust
love that you loved it. thanks so much for the kind words!! and yes, purz is great.
Thanks for another awesome episode! Gonna be playing with Suno all afternoon instead of working
each week, I spend more and more time with it. I think the charts thing was so smart of them because you can see meme-y things like the recipe songs pop up and try them on your own
see if you can top "It's Human Dog Food..." prob gonna be really hard
Have yall tried Udio yet? Tried to get on it last night but the queue times were too long. It looks insane!
Kebbin bannera!
Love the content! Purz was an excellent guest I'm to tempted to try stable diffusion now.
Dive in! It's semi-complicated but super fun.
WE HERE. WE HERE. WE HERE. Famous tomato-loving producer here and ready for today. Thanks for watching y'all.
Love this guest! Super useful information.
Glad you enjoyed it!
Another great episode! Very curious to hear an update on the self-driving stuff.
tried it a lil today. it wasn't bad.
Subscribed. I've been using computers since 1972, my first program was a basic demonstration of AI of a common program called "Animals", and I have been aware of neural network AI since the mid 1980s. Generative AI has taken a huge leap, but I don't know how some people that know how they work think that they are any kind of sentient, or ever will be.
Welcome! Yeah, it's a complicated question and I guess the biggest thing will be what happens at larger and larger scales. I think so much of the question of sentience is in part because we don't really know how we're sentient either?
I have a model 3 with FSD v12 and it is able to take me around my area with no interventions probably 90% of the time now. V12 does some crazy good maneuvers I never would’ve expected in the previous 2 years of testing FSD beta
ok, ok i will give it a shot today. i'm gonna attempt to go to a few different places with it.
Holy **** Kevin you came out fiery today. You feeling alright buddy?
Best comfyui explanation ever
Purz is the best.
I... kinda get it now. :)
I've been playing around it and TRYING to get my brain around it. But I agree Purz is very good at walking through it.
Cant wait til the new gov’t sketchers drop
same same same
Thanks for the call out - still want to do 6$ for a review rather than a lowly 5 star rating
Hahahaha ok well we will see how that goes. But yes we hear you.
I'm rate limited on GPT-4 and Opus most of the time since I have them pumping out so much synthetic data for model distillation and fine-tuning. The people who say it doesn't work are probably just inexperienced at prompting. There are some really neat insights gained though, like when distilling a model on synthetic text data from GPT-4, you get better results if you sequence the training to first use fake children's literature before phasing into more complex topics. Like teaching a student, it helps to arrange the content pedagogically. For synthetic data generation, it seems the key is to present a precise outline for the larger content structure and instructions for how it should express writing objectives before commencing on any piece. Without a human in the loop, the danger isn't so much that it devolves into inaccuracy, but rather it will lose sight of hierarchical structure and continuity of purpose, becoming redundant or overextending beyond a topical scope's definition. It's also helpful to embed notations that contextualize information metacognitively, using formal nomenclatures like Bloom's taxonomy from established curriculum writing, so that any content is presented mindfully as a learning experience. Inaccuracies aren't really a problem for anyone who knows how to employ tree-of-thoughts and if implementing breakpoints for iterative critical review and revision subsequences. The greater context length on Opus makes a lot of things possible now that weren't just a few weeks ago, but I still prefer GPT-4 (using a committee of experts) for most of the preparation tasks. For synthetic video, I'd guess a big focus is on procedurally generating cinematics in Unreal 5 and training on those.
omg this is so fascinating. i'd be super curious to see your prompts if you'd be willing to share one or two of them. like how do you start it off?
woah, yeah @AIForHumansShow just have this person on, I also want to know!
@@AIForHumansShow Here is a link to a starter template I made for a graduate level technical textbook drafting. Simple xml markup seems fine for interpretation. This should be relatively adaptable for other purposes, and once objectives are well defined, you can embed notations in the content for referencing them directly to improve revisions. Some links are included for my GPT review committee, but making those for yourself is one of the most interesting parts. Lowkey, Mythos is my favorite. I understand also that OpenAI has their own rubric and embedding notation for training student models, but I'm not caught up on the specifics. My larger aims later are to present knowledge domains or simulated spaces as networked graphs of text elements with multiple agents collaboratively navigating it for continuous section-wise iterative review and revision. For the simulated spaces, it would functionally be like a game of Zork where the NPCs build the world themselves from the inside out, but the text-based spatial descriptions can be formatted into image generation prompts, and later too adapted into procedural level generation for Unreal. I'm still brainstorming though! And really busy with school. Template link: www.dropbox.com/scl/fi/x93vq1y48p5lj9ezqscgy/educational-content-generation-template.docx?rlkey=tu6ne7yn4hblrqi8owomh4ppd&dl=0
If anyone is still into this, Karpathy turned me on to using Obsidian for document processing. It makes the workflow much easier. Highly recommended.
I've never heard the John Henry story with him being a coal miner. I've always only ever heard it with him pounding railroad spikes. Very interesting! BTW I was lucky enough to catch the last episode which was the first show? I think lol. Anyways, you know what's funny is I recognized Kevin immediately from when I was a kid. You've aged well Kev and your just as funny as I remember and more. Definitely subscribed! Great work fellas!
hahah well, I might've gotten it wrong but you get the idea. thanks for the sub!
Yep, same. I made a similar comment before seeing yours lol. Do some research Gavin! JK, the point was the same lol
Lol hey now I misremeber things all the time. Also regardless of what job he was doing, you got the moral of the story accross. But since your a celebrity were gonna chalk this up to a simple "mandella effect" 😅
@@DubStepKid801 Hahahaha me as 'celebrity' is as off as coal mining vs railroad spikes but I appreciate it. truthfully, I should've remembered it's one of my favorite American Myth stories.
The idea that you can choose specific areas to edit in AI is pretty awesome. But what really got me was the concept when Purzbeats mentioned we should ask AI if it wants to work with us...
yeah, ultimately it's gonna be a real different relationship (we hope it ends ok)
@@AIForHumansShow what's the AI equivalent of chocolates and flowers?
GPUS
When we all find out that you guys were ai the whole time that's gonna be quite a trip.
oh we've been clear about that FROM THE BEGINNING
I'm sure the AI will tell us ourselves soon enough
Udio is awesome and the queue times seem way better today! So hyped for ai music
Yeah it sure is fun to play around with - we missed it for this week because we record on Tuesdays but will def cover next week. Oooooh maybe EurovisAIon Song Contest?!?? Suno vs Udio??
@@AIForHumansShow haha that would be epic! Actually a song contest is a great idea, see who can come up with the wackiest most radical baja blastin tune!
finally went and played with it -- it's actually super duper fun
Kevin should have a red hat that says ketchup.
True Story incoming (this is Gavin)...
I love my Mustard hat so much that I nearly wore it out (the first one, this is the second one I have). It was a gift from my wife who knows how I love weird absurd dumb things to wear. My favorite shirt of all time was a white t-shirt that just said "THE MERGE" in bold red font. You get the idea.
My daughter thought she'd see what other hats were out there and lo and behold there WAS a red hat that says 'ketchup' in white lettering that normally would look very cool and fun and I thought WOW what a gift I love this.
I wore it out and got a ton of weird looks, include a few people that literally looked at me with what I can only describe as disgust. I didn't understand at all until I saw myself in the mirror briefly and realized it looked a LOT like another very famous red hat with white lettering.
Needless to say, I did not wear that Ketchup hat again but it still sits on my shelf beckoning me.
Gotta do a whole ARG just to leave a review on Spotify
hahahaha well we appreciate the effort whether or not it finds its way there
sorry but if you do it we send you a cookie
Google is a grower and TH-cam is a show-er… giggled the field mouse 🐭
😂👀😢
oh lord 😆😆😆
Everytime that AI See What you did There theme song comes on, I have to go listen to Square Hammer - Ghost. Weirdly similar chorus
not weird and now we shall also go listen
Just found out about your show. Thoroughly informed and well entertained. You have my subscription sirs.
Also, i kinda find it ironic that uber and others tried the robo taxi path and it might end up being taken by Tesla who solved rhe hardest part of the problem and figured the other side is easier so they could just add it on. I can see a whole bunch of orher scenarios where that plays out similarly. It'll just be held back by internal bureaucracy if others dont
First, thank you for watching and subscribing!
Second, yeah it's so funny to think that it was just going to be ubiquitous (and maybe ultimately it will be!) but I certainly hope we can hit it within the next five years.
We love it when new people pop in here and find us. Thanks so much for commenting!!
Did you hear about the controversy about AI art being used in the movie Late Night With The Devil? I only heard a little about it but supposedly people are calling for a boycott of the movie in protest. I was hoping to get your thoughts.
We did hear about this - we had it on our rundown a few weeks ago but decided not to talk about it mostly because we had a few other things pop up. Tbh, it felt overblown in the way that many of these AI stories are. It wasn’t even a lot, like a few particular images. There’s just a lot of anger amongst certain creative crowds around AI in general and we totally get it. But as we discuss in this weeks show, ultimately it’s a tool that can be used by human creativity to great effect.
HEY I MADE IT EARLY THIS TIME! Hope both of y'all are having a great day.
we love the plants!!
@@gavinpurcell now are you just saying that? Or have you actually watched my terrible content lol.
I'm just happy my continued engagement has been help y'all roll the boulder up the algorithmic mountain. You guys doubled from 3k in like what, a month?!
welcome. early or late, we're always happy to have you here.
love purz
💯💯💯
Call him Mr. Peebles if you’re nasty… how did that connection get made, Gavin?
hahahah look i've been around for a while
lol I replied from the show account but yeah... what they said
@@gavinpurcell I mean, the Janet Jackson song did instantly pop in my head, so thanks for that earworm...
Remember that time Tool did the exact same thing for a cookie recipe in the style of some Nazi dictator speech? Good times
Die Eier Von Satan
Ohne Eier!!! (Crowd Cheers)🙌
we knew you would know that
Sora the Explorer and her Large Language Monkey (Toots) need ayuda!!!
A, Yo estoy enseñando aquí!!!
John Henry set railroad spikes not mine coal iirc
alas you are absolutely right -- my fault (Gavin)
You need to check you the Jomboy Talking Baseball podcast about AI. It would be awsome to get those guys on what that thing about AI in sports!!!
will check it out!
Fuck yeah Purzzzz!
yeah he's amazing - so happy to have him on le show
The farmer's market has never been more BUMPIN! U kno cuz the... cuz the new speed bumps Robotaxi will b installing as it CAREENS down the isles
🤕
oh lord
I like it. Cruising for chicks FSD, I'm a foreign diplomat with full immunity FSD, I'm a 94 year old lady with bad eyesight FSD.........
is this the lyrics to a suno song?
We were promised Johnny Cabs
Gracias
You need to add your own names in the video description. C'mon, personal branding and SEO!!! AI not gonna do that for ya. Watched you, and loved you, but I have no idea how to tell other people who you are, as there are no presenter names in your description.
Hahah ok noted!
Names are: Gavin Purcell & Kevin Pereira
OK SEO DO YOUR THING
“AI artist” It's as ridiculous as saying I'm an airplane pilot when I take the plane 🤣🤣🤣
we kindly disagree. an artist is someone who works hard at their craft and feels something strongly about the work the put out into the world. ESPECIALLY what Purz is doing is that and isn't just machine in and machine out. Go watch one of this how-to videos and tell us what he's doing isn't specifically art technique that's something MUCH different than riding in a plane
An AI podcast that hates AI 👎
Huh? What do you mean?
@15m he's totality 100% right., but just scraping data isn't the problem imo., you are no more than a bunch of haphazardly thrown together algo sets yourselves, you scrape as you go cuz you can't absorb everything, AI can only record and react though, cuz no feelzbag hardware detected.... hooman gooputers lack lots of things., AI simply allows for those things without implanting chips & wifi in your brain.. but we need clones and a place to chill while we wait for them anyway,. AI is what will facilitate that too. soo FtheDs imo. i want immortality by 2042., get onn it
2042 seems actually doable but the question is... WHAT IS IMMORTALITY? just your thoughts / personality or full body immortality??
@@AIForHumansShow full on fresh meatsacks imo., We should all have a 25yr old clone for parts already ... thats the easy part,. so i see no reason why you couldn't do both and i would think both would become possible at the same time., cuz its going to be a convergence of all our tech like a singularity that blooms in to "stuff".., data storage ,neuro links, quantum computing, AI*, Crisper and the like,, all of it is coming to a head as we speak and soon 💥 all of the above and way more becomes possible.,. In fact what's on that list should be thoroughly explored immediately. we need to min/max this task & Elon being in charge of neurolink horrifies me..
@@oskarrecon8151 agree on the Elon thing but he really does seem like the only person (in public at least) who's pushing on this
@@gavinpurcell seems like it ,. but elon isn't even pushing for it., he's doing some other thing.. this is an A+B+7+*+j = immortal situation., only a few even know the thing they're working on might = immortality at the end of the day in 20 years,. everything from crypto to crisper might have a part to play
Another great show, I am always learning something and laughing every episode. Real quick...anybody else think that the 1897 Gambit photo looked a little like Kevin? Kevin how old are you...really?
...oh snap, I'm suing.
It's true he's at LEAST 75. He lives in a retirement home afterall.
AI Podcast: Powered by RedBull. A lot of RedBull.
Did we get just a lil *too* hyped?
@@gavinpurcell 🤣🤣 Maybe, just, 🤏...lol...(This was the very first video I watched this morning. As I was making coffee. Before the coffee kicked in. When everything else in the world is moving just a little too fast!)
@@therealking6202 It was the welcome, welcome, welcome wasn't it lol
@@gavinpurcell It was literally me, being an old man, waiting for my coffee, thinking, "These darn whipper snapping kids and their darn energy! Back in MY day...." 🤣🤣🤣
NUTRIENTS
yum yum yums
all the things you need to survive
Nutrient paste in a ubi is so kevin panera 18:10
TRUTH
Some really interesting research is coming out about using AI to review its own responses and then use that review to make better ones. Its called Agent prompting and researchers were able to make GPT 3.5 get results on par with GPT 4, albeit taking much longer per response. This can probably be used as a loop to create synthetic data. Use agents of your current models to create better responses, train new model on those responses, then use agents of your new model to create even better responses and repeat.
It really feels like if synthetic data is really gonna work, we're gonna go SO much faster.
Thanks again guys!😘
From now on, I need all my recipes in song form.