AI for Humans

Grok-2 Is Actually Good, Google's Huge AI Updates & More AI News

Join our Patreon: XAI surprised us all with the drop of Grok-2 and it’s actually very good, Google’s new Pixel event was AI heavy as Gemini Voice looks to steal Ope…

Show Notes

Join our Patreon: https://www.patreon.com/AIForHumansShow

XAI surprised us all with the drop of Grok-2 and it’s actually very good, Google’s new Pixel event was AI heavy as Gemini Voice looks to steal OpenAI’s thunder. OpenAI *did* drop a new model but it wasn’t the rumored Strawberry…at least not yet. Plus, AI crowds, way more Flux content and a special guest co-host Ben Relles who worked at YouTube forever and now in Reid Hoffman’s office!

Kevin will be returning soon - OR WILL HE?!? (just kidding, he will)

Join our TikTok @aiforhumansshow

And to contact or book us for speaking/consultation, please visit our website: https://www.aiforhumans.show/

// Show Links //

Grok (available on X Premium)

https://x.com/i/grok

Flux on XAI

https://x.com/bfl_ml/status/1823614223622062151

LMSYS Benchmarks

https://x.com/lmsysorg/status/1823599819551858830

Grok + Flux Examples

https://x.com/minchoi/status/1823698502909641144

Kermit’s Situation

https://x.com/AIForHumansShow/status/1823580442462957930

Google's GEMINI LIVE

https://www.theverge.com/2024/8/13/24219553/google-gemini-live-voice-chat-mode

Near Demo Fail

https://x.com/tsarnick/status/1823469426437710133

Pixel 9 Call Notes Feature

https://www.androidheadlines.com/2024/08/pixel-9-call-notes.html

Marques Brownlee on Pixel’s Add Me Feature

https://youtu.be/63EVXf_S4WQ?si=bpNkjlflbIi6ehgP

OpenAI Has Updated it’s GPT-4o Model

https://x.com/ChatGPTapp/status/1823109016223957387

The AI Crowds Controversy

https://abcnews.go.com/Politics/trump-falsely-claims-harris-campaign-ai-alter-photo/story?id=112776213

Wired’s Guide On How To Tell This Isn’t AI

https://www.wired.com/story/kamala-harris-rally-crowds-ai-trump-conspiracy/

Levelsio Using Flux To Create Models of Own Face

https://x.com/levelsio/status/1823199030199075277

Runway Gen-3 + Live Portrait = AI Liveblogger

https://x.com/EccentrismArt/status/1823059492520788342

Eleven Labs ASMR Voices

https://x.com/AIForHumansShow/status/1823046209294193020

Search ASMR in the ElevenLabs Voice Library

https://elevenlabs.io/app/voice-library

Reid Hoffman Meets His AI Twin (Reid AI)

https://youtu.be/rgD2gmwCS10?si=f4NBVQqGS7FYSXbE

Ben Relles on LinkedIn

https://www.linkedin.com/in/benrelles/

Real Creative (Ben’s Website)

https://realcreative.ai/

Transcript

Big shocker, Grok 2 has dropped, and it's good.

We'll tell you how Elon and the team at XAI somehow slipped near the top

of the leaderboards in the AI race and are integrating Flux to make

some, let's just say, some very weird AI images available for everyone.

Then, Google's huge new Pixel event where they unveiled and

shipped a new AI voice assistant.

Where's OpenAI in all this?

We're not sure yet.

We're really hoping to get advanced voice soon.

Sam, please.

Speaking of OpenAI, they did release a small update to

their flagship model, GPT 40.

Flux has updated a lot of stuff, and oh my god, I forgot,

Kevin can't be here this week.

We need a new co host.

What is this?

Where am I?

That's right, Ben Rellis is here on a new AI for Humans.

Okay.

. Welcome.

Welcome.

Welcome, everybody.

It is AI for Humans, your weekly guide to the wonderful world of generative AI.

We are here to demystify all the AI news and tools, and today, Kevin is not here.

He will be back next week, but I am joined by a new co host.

We are joined by Ben Rellis.

Welcome, Ben.

Hey, thanks for having me.

Love this show.

, and Ben, you and I have known each other for a very long time.

We go way, way back.

In fact, all the way back to the G4 days.

And tell us a little bit about what you're doing now in the AI space, because

you and I have been talking about AI for the last, what, two years now.

G4 Days, so that was like 2007 2008.

I was starting as a YouTube creator and it was, a big deal when your video

was featured on Attack of the Show.

So that's how we first met.

I was a YouTube creator from like 2007 to 2011.

Then I actually sold my channel as part of an acquisition to YouTube.

I was at YouTube from 2011 to 2021, and then I've spent the last few

years working with Reid Hoffman mostly focused on AI generated content, both

for companies that he's invested in.

But also some projects that actually feature read.

Cause he likes experimenting with all these tools like you do.

so read actually is doing this thing where he's got an AI avatar and you actually

brought me a question from read AI.

Is that right?

I did.

Yes.

This is a question to Gavin from read.

Ben, the funny thing about this is I would say almost every decision we made on

the show went against the grain, because we were putting the internet on TV at a

time where TV didn't trust the internet, we were trying to make, I wouldn't say

stars out of people on YouTube, but like, it was really, to us, the beginning

stages of something, and one of the things that we could talk about, we could

talk about this later in the show, is, Obviously, we're in another inflection

point as to how media is going to change.

I think right now the AI media space is very similar to 2006, 2008 YouTube.

And I think you and I can dive into that a little bit later.

But we're also going to hear more from Read AI later, which I'm excited about

before.

got more questions.

Yeah, but before we do that, let's jump into the news.

Okay, Ben, we had a big news week in AI this week, and probably nothing bigger

and more breaking than Grok 2, which was surprising to me because Grok is

XAI and Twitter slash X, whatever you want to call it now, their AI platform.

Elon has been talking a lot lately about how much money

he's spending on Grok training.

He has been bragging about the number of H100s that he

is going to put towards Grok.

And Grok 2 has come out overnight, and is actually really good.

It has come in at number three on the LMS training board, which is a

surprise to me because I don't know, Ben, if you had used Grok when it

first came out, but my experience with it was in general, not amazing.

Yeah, I used it and uh, I thought it was impressive in some areas

and maybe lacking in others.

It is interesting though.

Like I try all of the different image and video tools.

And then for the LLMs, I started with GPT 4 and I ended up

doing almost everything there.

So I can't say I gave it like, the full number of reps, maybe

to have a strong opinion on it.

I had a similar sort of experience.

In general, my problem with it was, which I always thought grok, if you're

going to use grok and you're going to integrate it directly into Twitter slash

X real time search, would it be such a useful thing that you could do with it?

Because there's no better real time engine, or at least there

wasn't, it may be a little worse now than what Twitter was, right?

It wasn't working that well before.

What's interesting about this now, I think more so than almost the real

time stuff is they've now dropped what is, I think could be a small issue.

Flux, which we've talked a lot about on the show over the last couple of

weeks Flux is a brand new open source AI image model from some of the team that

made stable diffusion, and it is making really good, really realistic images.

And now Flux has been integrated into Grok without some of the.

Let's just say specific.

I don't want to use the word censorship because it's not like these

companies are censoring, but it's

much more permissive.

Yeah.

It's much more permissive.

So there's been some really interesting images that have come out of Grok.

Min Choi, who does a great job of collecting a bunch of different sorts of

things, created some stuff where you can see what Flux, , is good at, it makes,

good text, there's a great picture that Dreaming Tulpa made where you see a

woman staring at the camera \ actually has embroidered words saying, follow

at Dreaming Tulpa um, uh, Images of George Washington, which we know have

been a problem for other image models.

I also am a little worried, though, because I, yesterday, used it to

generate an image of Kermit the Frog doing cocaine, which it did.

Didn't exactly have him with his nose in it, but there's a plate of white

powder in front of Kermit, and he is looking a little screwed up in some ways.

This is not necessarily about Grock, but I think the more interesting thing

to me about this is X slash Twitter is a mainstream platform, and now we

have, , a cutting edge image model.

Like the top of the line AI image model available to what

I would refer to as the masses.

Where does this go from here?

What do you think about this?

lot of things.

I think there's some areas of it where I'm like, this could

be incredible and so exciting.

We're going to see, the individual from his bedroom be able to create a full film.

And then there's also, of course, Kermit doing Coke that is nerve wracking and

how are we going to put guardrails around this and make sure that's not everywhere.

Probably with a number of the examples that we'll talk about the

show, that's the case where it's

like, As much as I'm excited about this being a incredible wave of like new

creativity put in the hands of people that wasn't there before it is also

both from the example you used, but also of course, like when you combine

that with video, something that starts to get really scary as you talk about,

an election year and everything else.

That said, I do feel like especially the last month has been the most exciting

since I've been tracking all this

stuff in terms of like how quickly you're starting to see things that

aren't just impressive because they're AI, but are also just really compelling

entertainment videos, et cetera.

It, what's interesting to me is there's a lot of people in the AI space who are

like, Oh, we've hit a bump or generative AI is going to not go much further.

And I will say, we've said this on the show before, but like clearly the powers

that be, including the CTOs of Microsoft and all these people outside of just like

you would refer to as the AI companies themselves are disagreeing with that.

They believe the scaling is going to keep going.

The thing I think that is those people, maybe the hardcore AI people don't

fully understand is that the vast majority of the world has not really

seen what these tools are capable of.

And I think video and audio specifically bring an entirely different space to that.

You know this from working at YouTube forever.

You saw, YouTube grow from what was like a pretty small service.

Never was like super small, but like into what is now the world's

largest, video company by far.

Do you see that kind of pathway happening now that these image and video tools

are getting available to everybody?

Yeah.

I think combination of like available and easier to use and with YouTube and

we both talked about, these analogies, but I think For so long, YouTube

was, billions of viewers, but still, most people weren't making videos.

on YouTube because it still took editing and final cut and all this stuff.

And then, for all of the different companies that had different, plays to

compete with YouTube, it was TikTok that really made like video creation so easy

that made that blow up.

And you could do it on your phone.

You didn't need to open up editing tools.

And I think there's a similar thing here.

There's like This incredible community of creators that are using these

tools, but it's still relatively small because most of them you can't

just open up and start, creating videos and see them minutes later.

And I do track a lot of these quick plug real creative.

ai put up a couple

Oh yeah.

So I try to track all these things.

What you start to find is that it's a lot of the same names over and over

again that are leaned in, Curious Refuge and Dave Clark and, Karen Chang and all

these people that are experimenting, but they're really good at it.

And so they can, have the patience and the skill to build something.

And I think part of the answer to your question is.

Not only is it about these tools being available, but it's like the easier they

get to use and the better they get, the more we'll start seeing like the volume

of videos that don't just feel like experiments that feel like, Oh,

this is actually like building an audience around a narrative series.

And I think distribution is a big part of that, right?

Which is why this is exciting and scary at the same time.

Like I think there's a little bit of, and this is not a diss

because I do this myself, the, 1000 monkeys can type Shakespeare.

That was always the interesting thing about YouTube in some ways is it

wasn't just about what was being made.

It was about the volume of that was, of what was being made so that there

will be things that, that climb to the top that are really good.

And I think that's probably the same case with this.

overlaps with YouTube a little bit in that I think early YouTube,

there was actually more narrative content in the top hundred channels.

You

had Freddie W, and Mystery Guitar Man, and all the sketch comedy.

And a lot of that got replaced, largely because of an algorithm change, by

individuals, first person content.

So I don't think that like, narrative content is going to overtake YouTube.

But to your point about the volume, as more and more people are starting

to figure out how to do animation, or put themselves in action sequences.

My hope is we actually see like more of that short form storytelling on

creator platforms, which, really isn't like a big part of the top

hundred thousand channels on YouTube.

Most of them are like very first person because that's what's practical to create,

That makes perfect sense.

And again, I think this is like where what's exciting

about the space right here.

Please like and subscribe this video on YouTube.

And always, forever, leave us five star reviews on all the podcast platforms.

We are available there.

And go to our Patreon.

\ we do have a Patreon right now that people are starting to drop a

little bit of a tip jar change into.

So let's move on to the other big news of this week, which is Google's pixel event.

And at Google's pixel event, there were a ton of AI updates.

They unveiled Gemini live, a brand new audio assistant kind of surprised me.

Their take on open AI is advanced voice.

And, you know, it's not.

Bad.

It actually does look like it's something that's pretty good.

I haven't tried it yet because I'm stuck in the iPhone ecosystem, but it is

available right now for a lot of people, which is different than advanced voice.

Supposedly advanced voice will be available for all chat

GPT plus users in September.

It's trickling out right now, but not everybody has it.

Now you can have a free flowing conversation with Gemini.

You can interrupt when you think of something important or change

topics as the conversation flows.

When I first go live with Gemini, there will be 10 different

voices for me to choose from.

Let's meet a few.

Great.

Let's get going.

Here's one of the voices I have.

I'm looking forward to discussing with you the world's most profound questions,

like why is pickleball so popular?

That is profound.

Uh, let's, let's try one more.

Or maybe you'd like to listen to a voice like this.

A pretty great one if I do say so myself, but don't worry,

there are more to explore.

Hi, Gemini.

How are you doing?

Hi there.

I'm doing well.

Thanks for asking.

It's always nice to hear from someone.

How can I help you today?

I think this is a really interesting step in the right direction for Google Gemini.

I assume, Ben, that you do not have a Pixel phone either, or do you?

I don't have a pixel phone, so I saw the demos, but I haven't used it yet.

The demos to me were, like, interesting and great.

, but in a lot of ways, this feels super useful, especially if you are in the

Google ecosystem, more so than some of the other stuff I've seen from Google lately.

Yeah, first of all, in terms of it not working seamlessly in the demo.

I've been there.

All the demos we're doing today are live, by the way.

So if I happen to come across this concert poster for Sabrina Carpenter,

I'll just open Gemini, take a photo,

and ask, Check my calendar and see if I'm free when she's coming

to San Francisco this year.

Gemini pulls relevant content from the image, connects with my calendar, and

gives me the information I'm looking for.

Oh, looks like we had a little demo issue.

Let me try one more time.

All

We used to do live demos at YouTube all the time, and I think they're

helpful and they almost always work.

But, sometimes they don't.

I've had it happen to me.

But in terms of the actual.

Once you saw what it was doing.

I thought it was really impressive.

To me, that is, the biggest hurdle for voice conversations.

Is it feeling like real time?

And I'm sure you've done these demos where you're like,

showing somebody a conversation.

But even if it's like a second and a half before

the answers, you're like, Oh, yeah.

It'll get to it and it just doesn't feel natural.

And so the, combination of that and multi modality and being able to feel like

it's watching something alongside with you, I think is another one of those.

It's a matter of, milliseconds, but it makes a really big difference in

how natural people feel like it is.

And it seemed like Gemini was like another big step in that direction.

Again, it's about shipping this thing.

That's really interesting because getting in people's

hands is going to be a big deal.

My issue in general with some of these AI tools is that they'll get shipped

and then people don't really fully understand how to use them or end up don't

making them part of their daily life.

I keep thinking that advanced voice mode, the clips that I've seen of it

once it rolls out to everybody that is really chat GT's, like secret

weapon because we've talked about pi,

So Ben has worked with Pi and Inflection before because Reid is a co founder there.

And what was always fascinating to me about Pi was they got to

that voice thing super early and I still think this is the killer app

Pi was sort of combination of voice, but also emotional intelligence.

So in having the conversation, you started to feel like it was

talking to you like a human.

I of course pushed myself to use it a lot because.

I was working with inflection, but once I did, after a few days, it did

become just like regular habit in the car, conversation for 10 minutes

or, with somebody and brainstorming.

I actually felt like once I learned how to ask questions, how to pause

it from talking all those things.

It became very practical and it was like one of the few

apps that I opened every day.

But yeah, to your point a lot of it was just feeling like it was a

natural conversation and , there's something that feels more natural to

me about having AirPods in and talking.

We should take a pause here to talk about the thing you've been playing around with

voice AIs because voice AIs obviously are super powerful You showed me something

that I was really impressed by You made a voice AI for your own grandmother, right?

Is that

I did.

Yeah.

After the fact, so my grandmother turned a hundred in April, I made a kind of

15 minute documentary about her life.

And then after I had made this, I shouldn't say documentary, little

film, photos, old videos, Yeah., as I'm editing it, I'm not a documentary

filmmaker, I'm like, oh, she didn't even say anything about, her dad, she

told this one story, but I need her like saying this one line about her dad, and I

could really use her, talking about this.

It was really helpful actually, use the Levin Labs.

And created the, voice clone of my grandmother, Eileen Chudnow.

And sure enough, about like 15 percent of this thing are lines.

I would call her up and say can you quick tell me something about Aunt Barb?

She would tell me something.

And then I could put that in the VO and that project is probably one of my

favorite projects I worked on this year,

because it actually, the AI tools did bring a lot of these old,

like photos and stories to life.

And we did some things with motion and we were able to expand photos.

So I thought it actually was.

A good way to help tell her life story.

Separate from that though, then we created, similar to the read AI thing,

a version using hour one where we can now have Graham Eileen talk about

the 76 ERs season AI I sent you.

I don't know if you can play a clip of her kind of like explaining AI.

But, it's don't know what the use case is yet, but definitely my family got

a kick seeing her walk through, her predictions for sports teams, artificial

intelligence, pop music, et cetera.

All right, let, yeah, let's play the clip where Ben's grandmother

surprised the rest of her family by using some words that you might not

normally hear the grandmother say.

consider this a W video from your grandma Eileen.

Some of you been asking what I have been up to.

Just beat COVID's ass.

Light work.

By the way, Eden, you're my least favorite great grandchild, my far eat shit loser.

Well, that's all from your old grandma Eileen in these 30 seconds.

Bless your heart, sweetie.

This is amazing.

We love Grandma Eileen.

She's amazing.

Tell her thank you for that.

I'll tell her I should clarify.

So read AI is built on All his books and speeches, that's really meant to

say things that the real read would say.

The Gram Eileen version is my nephew and nieces playing with having her say

different things, but it's not based on what the real Gram Eileen would say.

Just

Totally fair, and then just for our audience, for those of you who don't know,

the difference there is Grandma Eileen's voice was trained on Eleven Lab, so you

can make her say what she says in her voice, whereas Reed's stuff, you put in

all sorts of books and speeches into an LLM, trained it on that, and then had that

write the words for Reed, is that correct?

Exactly.

when he's asking you a question, he knows to not only like review your

LinkedIn page, but also to pull in things from Blitzscaling and review

and impromptu in his own writing.

A couple cool more Google AI things that came out of the Pixel event.

First of all, there was a really interesting call notes feature

that allows you to basically get your notes directly from a call.

It does have this weird voice where it says you're now being recorded.

But that's not that different than something in Zoom.

Even better, Marques Brownlee, who I know you and I both love did a really

interesting demo of the AdMe feature, which is a very cool thing that allows you

to take a picture of a group of people.

If you're the one taking the picture, You snap a photo and then you can actually

use AI to walk around, have somebody else take the picture and then put you in it.

Ben, you know Marquez for a while, but like, obviously we love

Marquez covering this content, but this feature is pretty cool.

It is a cool feature.

Yeah.

Also, Marques Brownlee, one of the YouTube GOATs, he's phenomenal.

I am conflicted about this one because on the one hand, I totally get it.

You want to remove somebody from a photo, add somebody to a photo.

But then I did see an interview with Kevin Rose, I think on a podcast.

And he was like, I would never use a feature like this.

Once you're

playing with a memory, it just feels like a slippery slope.

And he's not going to ever take somebody out of an image, put somebody into an

image with his own personal photos.

And I see both sides.

Part of me is like, well, yeah, with my four best friends on a mountain

alone, I want to get that great shot.

And then

part of me is like, that's weird to have a photo on your wall

that didn't actually happen.

I don't know where I fit on those two.

I really, and also I'm six foot five.

I used to like that people like give Ben the phone.

He can do the selfie.

He's got

the long arms.

Give it to Ben, he'll get us all in.

And now you don't even need me and my, selfie arms.

But no, I really like, there's so many things like this where part of

me feels like, yeah, that's great.

I could make my, grandmother tell stories about growing up.

And then part of me, it's like, That's weird to have somebody listen to a story

That was AI generated.

Just interview her and have her tell the story.

So I, yeah, I definitely see both sides.

I think I lean towards where Kevin Rose is where unless I like have to do it, I'm not

adding myself to a photo next to friends.

If it didn't happen.

Ben, I hate to tell you this is a generational thing and kids that grow

up with it are just going to use it and

have no problem whatsoever.

So we are going to have a weird world going forward.

We're going to move on now to our next story, which is OpenAI has released

a new model, secretly and silently, and then they talked about it.

It is not a real new model, it's an update to their flagship model, but the long

TLDR story is there was a relatively small Twitter account that got very big and

leaked some information that supposedly has to do with not only This update,

but then the GPT five slash GPT next, we don't know how much of that is true.

Supposedly this person says it's coming on Thursday, but

we believe this is all a troll.

Now the more important thing here, Ben, is that open AI has updated

their frontier model slightly with a better reasoning engine.

It seems like it is doing slightly better on math.

And some people are hinting that this is the pathway to what is

rumored as strawberry, which is a much better reasoning model built on LLMs

I know this is a complicated story and it's a little bit It's become a little

bit bigger than it seems to be right now where do you see open AI going?

there's some people who would say based on the grok updates or the

Gemini updates that open AI is falling back and that The moat that they had

created, which was really about how far they had advanced, maybe dissipating.

Where do you see open AI kind of in the current AI space?

I haven't been following all of the different strawberry teases.

I was traveling the last couple of days.

I'm not fully up to speed.

And I do feel like because it's open AI, people are always going

to get like more attention, more excitement about what's coming next.

But yeah, I think for me, like I said, it reminds me a little bit of maybe

like you go to Spotify and then that becomes your platform and it takes a

lot to move over to a different music service that's similar, that's been

like my experience with GPT 4, I'm just so used to using it that I get

very excited about updates because, you feel like you can immediately see, how

those changes impact your use with it.

So yeah, I don't know.

I'm I've been loyal to GPT for I try all the other stuff

and I guess everybody else.

I have no inside info.

Excited to see what's next.

Honestly, that is a point that I don't think is talked about enough.

And I think this is something you and I who are old enough to have been

through enough product cycles to see.

And this kind of goes back to the X thing.

part of it is where you spend your time, what is the place that

you feel most comfortable in?

Because a lot of people in the AI space are jumping from.

GPT four to Claude to many different pathways, like Lama, all these different

things, they're jumping back and forth, but honestly, I'm like you, like mostly

what I do, even though Claude is great, I don't right now pay for Claude, Kevin

plays for Claude, but I pay for chat, GPT plus, and I get most of what I need out

of GPT plus ultimately, it might be more of like a branding marketing thing, right?

If the moats get smaller and smaller now, my theory is that I think GPT

five or an extra, whatever it is probably not as far away as the

next levels of these next things.

So they will probably have a significant like advantage

when they drop that next thing.

But.

With Mark Zuckerberg talking about Llama 4 coming next year, obviously Elon going

hard on Grok, like it does feel more and more like these might become equalized

in terms of their intelligence level.

What is it gonna be like in a world where, you know, I always laugh about

the fact that apps are now advertised on CNN, like you see an advertisement

for an app because that's how you get attention for apps now, like you have to

advertise them to people on television.

Do you think we're entering a world where That's going to be the case.

There's going to be like five to seven different models and they're all similar.

And it's just going to be about who can garner the most attention.

Yeah, I mean, sometimes it's also like specific use cases.

I don't have the answers to all these, but if 1 is really good at

math, it might build an audience.

That's they're the best with math.

And if another 1 is really good at brainstorming, that finds a community the

same way that I don't know, maybe yeah.

Max appeal to creative people and, PCs, appeal to a different audience.

So I think there could be some of that where they build reputations,

not only through marketing, but they actually do things really well.

And then, yeah, there's so many analogies of AI that I try to catch myself.

I felt I went to some of these AI conferences and everybody talked about

like the comparison to photography and how, AI is similar because people

were scared of photography and then it ended up being this wonderful thing.

And.

One analogy maybe here is that to me, the iPhone updates at

some point were like overkill.

The,

The camera, it was fine.

I could stay with an iPhone 12 for three years and not feel like I got

that much more with the next phone.

And I do wonder with some of these updates, if for some people like,

Oh my gosh, I can plan my whole business now with this, this is.

But for the average user, they're like, yeah, I can still sort of like ask for

suggestions for what to do in Arizona.

And it's pretty similar as it

was before.

Think the other side of this is, we talk a lot about AGI in the show,

this idea that all these companies are shooting for an artificial general

intelligence that can do a lot of stuff.

And It might just be that like there's going to be like a very high level

machine learning intelligence that will solve the big problems, right?

Which like in the dream world, there's a you set, you send an AI off for a month

and it comes back with something new from physics and then we could take that and

break it down all these different ways.

That is not the way that we're currently using AI.

So it might just be that like for the masses, the consumers, AI will just

trickle out and get better and better.

And these larger models will be used for the big problems in some form or not.

totally.

And it's so hard to predict years down the road, what is this going to look like?

And I'm actually put together like a breakfast in LA where it was Reed and

like six of the top creatives in AI.

You would know a lot of them.

It was like

Don Allen and Terrence Southern.

And and I asked at the table, like, where does everybody think

this is going to be in five years?

And, some people answered and then that same night read was on a panel with

JJ Abrams and he like said something.

I'm like, I'm not going to ask the dumb question.

Where is this going to be in 5 years?

Because nobody could ever possibly know.

I was like, was that aimed at me?

I just asked that this morning and then, and then 2 days

later, he was interviewing.

He's I'm not going to ask the dumb question.

Where is this going to be in 5 years?

So anyways, I told him that story.

But it's like I do think there's an element of.

It's so hard to, cause you were asking like, where do you think

this is going to go to really have an understanding, like longer term,

what this is going to look like,

There's another story that kind of ties into this, which is,

I don't talk about politics a lot, but it is political season.

I think it's important to discuss this conversation around the

crowds issue that has come up.

There are two candidates in the presidential race and of course, One

of them has had a surge recently.

Kamala, there was a change in the presidential candidates and Kamala Harris

saw that.

yeah

yeah, and Donald Trump has basically said that the crowds that she is drawing in

one specific event were actual AI and, and, and Ben, I don't know if it's as

important to talk about what Trump has said here, but more about the idea of how

we tell people the images are not AI and think in different ways to look at it.

Wired wrote a really good article, which I hope to see more of, which

is basically giving people the educative tools to understand when

something is AI and what it is not.

And one of the most interesting things about this is, do

multiple angles of this exist?

Can you see multiple places?

And I think that's an important thing to come across.

We should just be clear, it wasn't AI in this instance, but

Totally.

I was going to say I'm not a newscaster, but that case, 100 percent was not AI.

There's video, there's other angles, not AI.

Yeah.

But it does open this conversation about deep fakes are not just about

the times when they are actually faking information, it becomes a situation where

it changes the ability for people to say, oh, that was a deep fake if it wasn't.

And I think that's a weird world that we're entering.

I know you speak and talk a lot about AI and obviously working with Reed

must think about this stuff a lot.

How do you educate people to understand this stuff so that specifically they

don't get caught in a loop of believing things that aren't necessarily true?

I can't say that's exactly my role, educating people on how they can not

get caught in a loop, but that would be a good thing for me to learn.

That's, that sounds like a smart thing for me to be able to do.

I think that.

I thought the obvious way that AI could potentially be used in a negative

way would be to create an image of something that didn't happen, it goes

viral, and everybody thought, oh, this candidate was, smoking a cigarette

with so and so, but actually, AI generated, right?

And that might happen.

And then the other use case is this instance where it's like, something

really did happen, but because AI imagery is getting so good, you can

credibly put out something that says the crowd wasn't that big, which is like,

you know, big issue to Donald Trump.

So that once it's out there, it puts that seed in people like, does

you really have those crowd sizes or whatever it is, and it could be

something more serious than that.

And it's really tough as a content creator to even know because I saw

like one video that was, can you believe what this guy is doing?

He put up an AI generated image.

Here's how we prove it.

But it doesn't necessarily mean that person even knew they were using an AI

generated image.

So there does, yeah, really, Necessitate a need for like how as

quickly as possible can people verify whether an image is real or not.

And there's a lot of companies working hard to be the solution to that.

And honestly, I think this is where, and not that I trust necessarily

what's going on at X from this side, but I do believe social media can

actually help in this way in some form.

And there's a lot of arguments that like, obviously.

It can dissuade people and it can do stuff.

But part of it is about how much signal you can get on something,

? Like in general, I have seen people be very good on signal on things

like this pretty quickly and pretty honestly, because if somebody is

dishonestly trying to say something, it does come across relatively fast.

And this is where you can percentage wise determine how many people are

saying something is real versus not.

There's like a certain number of yeah, platforms where it's a

scrolling feed and I don't know, it's tough to know how horrifying

one individual user's experiences.

Mine was saw some weird thing about was, that rally, a fake image.

In front of the plane, but then two videos later, it was like disproved.

And then I was like, I had seen enough in it, but somebody else's feed

could just be like video after video,

convincing them that like the whole, campaign momentum is a mirage and

not happening and it's so hard.

And I think about that with my own kids too.

And I don't love the scrolling platforms for that reason,

because.

It can be a really great experience for some people, but for other people,

the same platform can be a nightmare.

So that to me, to your point about like social media being helpful, I think

some platforms are better than others.

guess is if you're scrolling through videos, you can't count on the

videos correcting something that was misinformation seven videos earlier.

Yeah.

No, that actually, that makes a lot of sense and it is something

to be really worried about.

Time to look at some of the stuff we saw on AI this week that we weren't

able to try, but we want to shout out.

It's really cool.

It's time for AI.

See what you did there.

Sometimes you're scrollin without a care, Then suddenly you stop and shout.

Hey, I see what you did there.

So Ben, one of the most interesting things that has come out in the

last couple of weeks is flux.

As we talked about with the Grok update, flux is a new open

source AI model for imaging.

And it is doing some really interesting stuff.

Levioso, who is a very interesting follow on X, has created what a

lot of people have now done, which is a Laura of their own face.

And he has been able to drop himself into all sorts of pictures to do stuff with,

and to talk about some of the things you discussed with, you know, Marquez's take

on Admi, or talking about Admi earlier.

Now you have photorealistic versions of yourself that can be anywhere early on

with stable diffusion, people were making apps that did this, but now it is going

to be very easy to do this for yourself.

And to the point that we made about Grok earlier, Grok could

conceivably roll this out.

It's a very cool thing.

It's super fun to play with them.

But again, it does feel like there's a couple of images that Levioso makes

with like him putting his arm around Donald Trump or his face as Donald Trump.

And like, it does cross that line a little bit sometimes is like, well,

you could put people into something, but did that actually happen?

Or what was it like?

. I know.

I know.

My all time favorite photo of myself.

Okay.

Not including family photos.

I do have a photo of me and President Obama.

We're shaking hands.

We're both smiling.

Now anybody can have that photo.

And that is like, you know, the moment that I'm so glad it's captured on film.

And then it is very bizarre with video, not just photos.

I work with a guy named Parth.

He created

the LLM for Read AI.

Okay.

And um, he was just over yesterday and showing me a video where he took

one image of my face, put it on him.

He does a pretty good Ben impression.

I have to say I can send you the video if you

can put it in here.

But um, it was a little surreal.

I've seen those kinds of demos before from metaphysic and deep voodoo, but having it

done with your own face on somebody else, that's a friend of yours it's bizarre.

And it's interesting you mentioned Metaphysic and Deep Voodoo

who are both like large tech first companies that do this.

The thing that's crazy to me is the fact that it can be done off the shelf.

So actually, as you mentioned that, what's interesting, Partha's probably

using Live Portrait, which we've talked about on the show, which is an

interesting, plugin that allows you to use your face to act out stuff.

Well, Eccentrism Art actually used Live Portrait plus , Runway Gen 3,

to create what is a live blog looking thing of a woman talking to camera.

Hi there!

Okay, this is gonna sound really strange, but I had the weirdest dream last night.

Of course this has some editing involved in it, but Ben, when you're

watching this, what are your first reactions to what this looks like?

You can see like they try to get the nods and like trying to get

her to like, kind of be the thing.

Yeah.

Yeah, I haven't seen this.

What's so bizarre about this is lonely girl 15

was the first ever breakout series on YouTube, right?

And part of what made it such a wild story is that, lonely girl 15 was fake.

I still think it's one of the most innovative things

ever done with online video,

even though

I honestly with video period, I think cause it was

video period.

It was so smart.

It's a lot of what I based, the Obama girl series of videos that

I did was based on lonely girl 15

originally was going to be called Obama girl 15.

And she's going to be blogging about Obama,

but anyways, I just, I haven't seen this clip, but what's so interesting about it,

it definitely gave me like lonely girl 15 vibes and whereas that one was fake

as in like scripted and a real actress.

Playing the part of a blogger.

This is like next level, like lonely girl, 15, have a fake blogger,

create a whole story around them.

And yeah, it's, just, it's pretty, I, of course, I've been thinking

a lot about what does this mean for the creator community

when suddenly.

You're going to be able to have creators that might have that parasocial connection

and don't exist.

And what's, how's that going to play out?

But yeah, I hadn't seen a demo yet like that of a vlogger.

It's interesting because it just combines all the tools, right?

And I did a video on Monday, which was using the Flux Realism

Laura, which is what this is using to get the initial image.

So Flux Realism Laura is a way to use Flux that is trained on very realistic faces.

You've probably seen all those pictures of holding up like

Reddit I'm not real sort of cards.

It's really crazy.

I do think we're crossing into something pretty bonkers now though, right?

Because image to video tools have gotten really good.

You can see how they can make it a straightforward and then it just

becomes an editing issue, right?

Do you have a talented enough editor to make it feel right?

And as we know with YouTube, we've had 10 years of a generation

getting good at editing, right?

This is something where like editing is something that's like.

It's funny.

I was talking to my wife, who's a novelist the other day.

And she teaches writing to kids and tries to get people like, excited

about writing at a young age.

And I think kids almost editing is, is in that first skill set now, which is

Yeah, for

thing to think about.

Where it's like writing, drawing, and now editing.

That's a thing that people actually do at like five, which is a crazy

Yeah.

And prompting.

It's

interesting that if you can prompt these things the right way, you could

get someone to just vlog for 24 hours straight about all these different topics.

And, then have, another AI model pull out the best hot takes

and turn those into videos.

So yeah, it'll be wild to see where that goes

One other thing I wanted to point out is there was a great post

from one of our favorite ex users.

Her name is venture twins.

She found out that chat GPT can determine how tall people are with within an inch.

This is something I hear from my daughters.

My daughters are, 19 they talk about how.

Boys, including me, they try to explain me always lie about their

height and I am actually 5'11 and a half and my daughter's, I used to

call myself six foot and my daughter's giving me so much crap for it.

Anyway Justine was able to find out that chat GPT can, if you give it pictures

and put it in proximity, can tell how tall somebody is within an inch.

So like this is real world use case of AI

It's funny, this is the second time in this podcast I'm mentioning that I'm

6'5 so it seems like something that I work into every conversation when

I really don't, but, I'm 6'5 and yeah, it's funny after COVID,

I'd meet people all the time, and they're like, I had no idea you

were gigantic, because I spent two

years, on Zoom.

But to your point before, which again, this is people are talking about a

lot like, It's, it's very interesting that you could be conversing with

somebody over video and then when you meet them, not realize that

they looked, 25 percent better this,

All that stuff that like, yeah, not only is it like AI images, but it's

just like everyday conversations.

I don't know.

It feels like again, there needs to be like certain like societal norms

around don't change, your image if you're just facetiming with

your girlfriend, whatever.

Yeah.

I think it's going to be, IRL meetups are going to be much stranger because

you're going to have different looks at different people and everything.

And in some form, it's going to just get weirder.

All right we should talk a little bit about some of the weird

stuff we did with AI this week.

This is what we do and talk about the stuff we did get hands on with.

As I mentioned, I did go and play around with the phone.

Flux realism, Laura, please.

You can check out our YouTube video on that.

But the weirder thing I did.

Is 11 labs.

I was on 11 labs because of this.

I was trying to get a voice to use for this video.

And I saw that on 11 labs, there are now ASMR voices, Ben, which seems to

me to be a very bad idea, but maybe not because the ASMR people are out there.

So if you listen to this, just, I'm going to play this link here.

I just did something very quick and I just want to try and see what happened.

And it's pretty shocking.

This will not go well, not for you, not for us, nor for any of us.

Click.

Bruh.

Woah.

It's ASMR.

It's basically what those people are, have done for years.

And this woman maybe smartly created a very good model for it.

And this feels like a place then where AI could replace a lot of

people doing this thing, I assume.

Yes.

Sorry that was looping.

I couldn't figure out how to close the video so it was looping in

my head while you were talking.

Yeah, and it's also bizarre that you're the one that said it, so I'm like,

is this Gavin whispering in my ear?

Just in a different voice?

You know what I mean?

Like with the AirPod

Ben!

I don't know if exactly, if you're like, it's Gavin.

Yeah.

Trying to think as as I try, there's always a good use of this stuff, but yeah.

As MRI mean, I never I never really went down the rabbit hole of the ASMR videos.

But if you're into doing them, it does seem this could be a way to have

like ASMR about whatever you like.

I like fantasy baseball.

Oh my god, that's a great idea!

The ASMR fantasy baseball, if you could leave your fantasy lineups,

The New York Yankees Aaron Judge today had an amazing game.

Two home runs, unfortunately, your pitches weren't as great, but it's okay,

I'm sure you'll do better tomorrow.

it seems to me like hyper personalization

is one aspect of AI

that, is unpredictable and I'm sure will play out in fascinating

all right.

So that was what I did with weird ASMR AI this week.

Ben, what have you been playing with, with AI?

This morning I was playing with read AI and having it generate a few questions for

you, co host of the AI for Humans podcast.

I can have you play a couple of those and then maybe I can explain

where I think, this has potential.

Looking back at your time on The Tonight Show, which segment or episode

stands out as a turning point for you in terms of realizing the power

of the content you were creating?

Do you see any parallels with the kind of content AI is making possible to

create now, or could in the future?

Okay, so this is a good question about the Tonight Show.

I think in the Tonight Show times, and really in the late night times, I think

the thing that really made me feel like we were onto something different

was the late night hashtag segment, which was a way to use, at the time,

Twitter, To help start and generate and continue making a conversation with

the audience into a television segment.

One more question.

I took a look at your impressive LinkedIn page.

Great banner art, by the way.

I saw you were an executive producer at NBCUniversal.

What's the craziest idea you pushed that got greenlit?

How did that experience shape your view on AI's impact on traditional media?

Oh, okay.

This is interesting.

So clearly read AI has seen my LinkedIn.

And it's, amazing banner that I've got there.

So I appreciate that.

I'm this is a tricky question because one of the funny things about this, and

this might be an interesting thing for us to talk about is it says EP at NBC

universal and because I had to put it on LinkedIn in some form or another, it

encompasses so many different things I worked on, but it is, it shows in some

ways the slight limitations of what it's like to have to read a piece of material

and then generate questions from it.

Do you know what I mean?

Because,

Totally.

Oh, for sure.

yeah.

And I think that's something maybe to dive in on this.

So tell us a little bit about how.

This thing is made for Reed.

And if you're, if, again, if you're not familiar, we're talking about

Reed Hoffman, the former founder of LinkedIn now is very active in a bunch

of investments and Ben works for, as we talked about at the top of the show.

Just an experiment at this stage, but we were already playing around

with doing a video version of read partner with a company called our one.

And we had done his voice with 11 labs.

And then right around that time, we were tinkering with this opening.

I came out with custom GPTs.

And so then like in a matter of 48 hours, we were able to start putting a

lot of reads books, speeches, podcasts.

Into GPT to be able to have read, give answers in a very read like

manner that GPT wouldn't both in terms of the content itself.

So in some of those questions, he's like referencing blitzscaling

and some of his own thinking, but also in the way that he speaks.

He might have phrases.

He says a lot.

I, I noticed when I listened to myself recently, I'm like, I

say, makes sense all the time.

Why do

say uh, uh,

ubiquitous, ubiquitous.

And uh, YouTubers will love to comment on.

Yeah, I had a friend of mine, a comedian was like, the first time you use a

word like ubiquitous, he sounds smart.

The second time it's like, oh, he really just thinks that's a great word.

And I, yeah but similar, right?

If Reed tends to use, a word like brilliant instead of smart,

it should pick up on that.

And so I think that's what made this unique is that, as we're building all

of these different use cases, we've used read AI to, review business plans

at Stanford and give video feedback.

He's done interviews with, Bloomberg and wall street journal, and he's

done, speeches where we translate it to 10 different languages.

All of them were trying to figure out like, how can this be additive?

How can it, do something maybe interesting that you couldn't do with video before.

And so it's combining, yeah.

The audio, the video.

And then Parth, I mentioned before, created this LLM, every, you know,

couple of weeks, it's getting better and smarter and more knowledgeable

about what Reid said in the past.

So yeah, that's the concept behind it.

And then with LinkedIn specifically, it can review your LinkedIn page.

And to your point, sometimes it might pick up on something like artwork, but

other times, it could be limiting and how it thinks it should take a single bullet

point and turn that into a question.

Does this become, I think it's great that Reed's doing this

because I think it opens the door.

First of all, it shows people what's possible.

And in it, in what we were talking about before, even with deepfakes a

little bit, it shows people like, oh, this is something you can both create

for good, but also people could have a version of this that is something that

you have to be aware of what it is.

Is it going to be productized in some way?

What is the use case?

Like, why would people pay for something like this?

Yeah.

I think people will productize it.

I don't think that's what we're looking to do with

this, but initially we created it.

There was this idea of wouldn't it be great to have, feedback

from someone like read as a mentor in your pocket, pull it up.

You're thinking about doing something with AI for humans.

Yeah.

And you can get his perspective and then you can get Steve Jobs perspective and you

get somebody else and you sort of have, different thinking from different types

of, mentors, thought leaders, et cetera.

Or it could be advice from, you know, your dad or your best friend or

whoever you trust.

Reid, you mentioned this is about exploring what's possible, that is

very much, what Reid is looking to do.

Show and tell, he wrote a book with GPT called Fireside Chatbots,

he did this project, he's always sort of experimenting, To

figure out where this is going.

You need to actually use the tools and figure out, and he's on all

of these things all the time.

And I try to do the same thing largely because that's how you

can see where things are headed.

But in terms of like how this will be productized.

Yeah, I think.

Informational content and educational content makes a lot of sense for

these types of digital twins.

I don't think it makes a lot of sense for, a creator who's

opening themselves up to their

community

and being themselves

You want to have a connection with a real

there.

I think you want the real connection.

And then there's other cases like my, I have twins that are 16 and

I'm like, surely we don't need to be paying for an essay to tutor.

There

could be somebody who's watching them, talking to them, explaining how to do

this stuff in real time, interactive.

I'd rather pay for that.

Not that I want to put an essay to tutors out of work, but then they can

do it at all hours and they can, it's customized for them and all that.

So I, yeah, it's a bit of both.

And, it's funny, another one, cause we talked about the grandma lean

video I did.

And on the one hand, I think it was funny to see her, break down

Paul George coming to the 76ers.

It's just funny to watch.

But then on the other hand, to make this video, I interviewed

her for two and a half hours.

And I had like, a lump in my throat the whole two and a half hours.

I was like, I can't believe I haven't had this conversation

before, and she's a hundred.

This woman is remarkable.

Like, I didn't have that.

And I'm much happier that I have that interview than I'm,

you know, Excited that we can make her

rethink, how she would have said

something.

And so on the one hand fascinating that, yeah, people probably will

preserve versions of their loved ones that they can interact with.

But then on the other hand, the thing we could do in 1986, I think people

should do much more of, which is just sit down with the people you love.

Interview them, hear their life stories and have that video preserved.

And there probably will be things with AI that you can then,

ask it questions or whatever.

But like, actually what made that project cool is not

Not

the

I can now,

the connection you

the AI.

Yeah.

I mean, Not to be hokey about it.

I know this is AI for humans,

but,

No, it's the human part.

That It's a human part.

That's

oh, that's true.

Yeah.

And it was funny actually in the early, early in the interview, she's like,

And have you heard about this AI?

Oh my gosh, the things they're doing.

That's what people should leave with this year.

Just go interview the old people in your life so you

get them

on camera.

I had a wild experience.

Actually.

My grandfather died in 1998.

He lost all his brothers and sisters in the Holocaust.

Never talked to him about it

really.

Cause I was young when he

died.

And then um, in like 2005, I was basically Googling myself

and I stumbled into basically 10 hours of interviews with him about the Holocaust

way.

Wow.

it was.

Kind of wild.

Nobody in my family knew it existed.

We had seen like three quotes from this interview in a book, but then

there was like 10 hours of audio tapes

of him recounting the whole experience.

And I did use 11 labs to recreate his voice.

I actually want to do this.

Cause I think this could be an interesting thing where.

It's 10 hours.

Not a lot of people want to sit through 10

hours of him recounting the Holocaust, but you would be able to say, retell

the story in five minutes at a seventh grade level, there's ways to maybe take

this like incredible artifact of him recounting his story of the Holocaust.

And, make it like more, I don't know, accessible for audiences.

But yeah, that stuff is priceless.

And I do, it's funny because part of me feels like a lot of bloggers

are probably now in regular jobs

and it was a bizarre thing to go through.

But if nothing else, they, a lot of them do have this like

incredible time capsule of what they were going through.

Yeah.

Ben, thank you so much for being here.

Where can people find you online?

And what should they check out?

People can find me.

LinkedIn.

This is the first time I've co hosted a podcast.

You're doing

great.

You did a

video a little more.

I appreciate it.

I want to start doing video more, but, yeah, really, it's linkedIn posts.

So let's go LinkedIn, Ben Rellis.

I have a website, real creative.

ai for

something to check out.

We try to curate these different examples of cool AI creative projects.

And then if you go to the top, you can store it by project type.

And we have over 300 creators curated.

I'm just like fascinated by this community of AI creators that are like.

They're not really as focused on like views and money right now.

They're focused on

making cool

experimentation.

I'm sure later they'll want the views and the money more too.

I would say, okay, this is an odd one.

I would say in a little bit of a non sequitur, I would check out some of

the tributes and some of the things being written about YouTube CEO

Oh man.

camp.

I get choked up.

All right,

okay.

Non separate or something to check out.

I would say read some of the tributes to YouTube's um, CEO, former CEO uh,

Susan, who passed away last week.

She was the CEO when I was there for most of my time there.

Remarkable woman.

I admire her so much.

So it's not an AI thing, but it's been fascinating being on LinkedIn because

I just like, I'm reading all these tributes from people that I was really

close with when I worked there and I don't know, it was like Obviously

heartbreaking, but also inspiring to read all these experiences that people had.

So for people in the tech industry that want to read about, you know,

an icon uh, yeah, look up some of those tributes to Susan and what

she was able to do with her life.

She sounds like an incredible person and how lucky you were

to be able to work with her.

And what she did for YouTube is pretty remarkable overall,

And also not just how big YouTube got, but I have to say it was a big

reason I stayed there for so long the culture of that place people really

admired her, and so, like, yeah, in addition to, of course the insane

growth it had over that time period.

I just think she like really was passionate about the people at YouTube,

the creators, and created a culture that made you want to be a part of it.

That's amazing.

Ben, thank you so much for coming on this week.

We really appreciate it.

And Kevin, we miss you.

We'll see you back again next time.

But that is AI for Humans, everybody.

We will see y'all next week.

And again, Ben, thank you for being here.

Yeah.

Thanks for having me, Gavin.