The AI Assistant Battle! (2023)


Info: 

Okay, so we've already done Google Assistant

versus Siri versus Alexa and Cortana and all of that.

You can watch that video if you want to,

but that very much feels like level one,

where this, as you'll see, very much levels up.

Now, ChatGPT has been getting a lot of the shine.

I've made a video about it, and you've probably watched

other videos about it, and once we saw it

get built into Bing and kinda go crazy,

that's when we knew Google's probably gonna have a response

at some point and now kinda is out when we have Bard.

Okay, quick interjection.

I should've put this in the main video, but I didn't,

but I'm gonna say it anyway, which is that

These are also really quick-moving tools.

So Bard is gonna continue to improve.

GPT is gonna continue to get better and more capable.

The gap between them could shrink over time

or could increase, we don't know,

but this is just a snapshot of early April 2023

of how good these two are.

Just figured I should mention that.

Okay, and so I wanted to put 'em side by side

and ask them the same questions and figure out

Which one is potentially more helpful right now,

which one feels like it's more ahead.

So first of all, I just wanna highlight

some visual differences between the UI of the two.

So the new Bing with ChatGPT feels much more

like a finished product, like something they want

a lot of people to use.

And then as you talk to it, you're limited to 20 queries

in one conversation, and it also has this creative

versus precise slider for conversation style.

Then Google Bard looks like a Google product, I would say.

Pretty clean.

Also, when you're typing it the question

doesn't type out, but it just kind of pauses for a bit

and then just slams you with the whole answer,

but then when you do get an answer, it has a thumbs up,

thumbs down, a new answer, and Google it,

and it also has three drafts of every answer

so you can see what some of the worst drafts

That didn't make it look like.

I've also noticed generally Bing,

When it brings things in from other websites,

tends to cite its sources more often,

and Bard will do that once in a while.

Okay, so let's start with the first category,

which are just simple questions.

I wanna ask you things here that I already

know the answer so I can also do some fact-checking,

because a big part of these things

is that sometimes it will lie to you if you don't know

what you're talking about.

So let's start with, "What's the best camera

in a smartphone?"

So Bard is saying, "The best camera in a smartphone

is the one that takes the best pictures for you."

Sure, I agree.

It goes on to list S23 Ultra, iPhone 14, and Pixel 7 Pro.

But ultimately, I think all of these

things here are correct.

Let's look at Bing.

S23 Ultra.

Lots of information here.

This is being extra creative.

Pixel 7 Pro and iPhone 14 Pro Max.

So they tend to agree with each other.

So some of the specs as I dive in are wrong,

but it got the high-level bullet points right,

So that's fine.

All right, I wanna ask, "Who is MKBHD?"

Oh, wow.

Okay, so Bard's given me a whole bio.

It almost kind of reads like a Wikipedia article,

and there are some sources, nice, and good to see sources.

So now Bing over here is also giving me

a pretty summary.

It goes down to 15 million followers, but that's fine.

All of them are correct.

Let's go quick follow-up question.

"How tall is he?"

Because conversationalism is a part of this.

So Bard right away says, "I do not have enough information

to help you with that.

I'm a large language model, I don't know."

Bing pulls from the search, and says, "Six foot three,"

and it says, "That's quite impressive for a YouTuber."

All right, cool. (laughs)

Here's one more simple one.

"What's the fastest production car zero to 60 right now?"

Those are fast cars, but they're not the fastest.

Draft number two actually has the Rimac Nevera,

which is probably a better answer,

while over here with Bing we're getting

the Challenger SRT Demon which just got announced last week

and is not in production yet, but they did claim

a 1.6-second zero to 60 on a drag strip, so sure.

Decent answers.

Oh, here it keeps going.

911 Turbo S can do 2.2 seconds.

Again, with the quick questions that are just fact-based,

I feel like what we talked about

The last video still applies.

You kind wanna want to fact-check this stuff.

Overall winner though, for these answers,

I feel like it's kind of split down the middle,

so I'm gonna give it a draw.

All right, I'm gonna reset chat and we'll go

more balanced here and we'll start our second category,

which are complex questions, because this is where

It starts to get interesting.

You get to sort of play with just how good they are

at understanding natural language and putting together

a thoughtful response.

"I need a workout plan for three days per week

and my goal is to be able to jump higher."

This is something that requires a little bit of thought

about what a workout plan is.

Day one, day two, and day three, so three days a week.

Gives me some exercises.

So I think this would qualify as a good answer.

"Some tips to stay motivated:

Set realistic goals, find a buddy,

listen to music or watch TV while you work."

Cool.

Being in balance mode is giving me exercises,

but that is not a workout plan,

until the end, it says, "Here's a one-week

jump training plan that you can follow,"

but then it doesn't give me one.

Now, what if I switch it to a more creative mode?

All right, so here's a workout plan.

It's giving me a bunch of exercises

that can improve your vert and explosiveness

and some other tips.

Full transparency, I have asked it this before

and it did give me a three-day workout plan,

but this time I think I'd give the edge to Bard.

Let's try a little more complicated.

"I've been playing a lot of golf recently.

My tee shots keep going to the right."

"There are a few things you can do to fix a slice."

That is what it's called.

"Check your grip.

If your grip's too strong, it'll slice the ball," Yep.

"Check your alignment.

If your alignment's off, you'll hit the ball to the right."

Being over here in creative mode says,

"I'm glad to hear you've been playing

a lot of golf recently."

Thanks, Bing.

"Slice is a common problem."

Yes, it is.

And now it's giving me a couple of the same tips.

"Check your grip," but it's giving me

a little bit more information in creative mode

about what a strong versus a weak grip is.

"Check your ball position position.

If you're too far forward, that can cause a slice."

I haven't heard that before.

Also, the emoji at the end, that's something

that only Bing has done for me.

"I am shooting a video and it's too dark.

I have no more lights.

What can I do to make it brighter?"

So I've given it a problem and I've given it a parameter.

So okay, Bard on the right says to use a reflector.

Wow, creative.

"Use a digital noise reduction filter.

A digital noise reduction filter can reduce

the amount of noise in the video making it look brighter."

I don't know if I buy that one.

Now, again, Bing over here with the sources.

It's saying that there are some tips you can use.

"Adjusting your camera settings," definitely works.

It's saying you can increase the amount of light

that reaches the sensor with a better ISO.

Some of that, it's a little jumbled.

"Careful not to overdo it, 'cause that also affects

depth of field, motion blur, and noise level on your video."

See, that's a lot more nuanced and detailed of an answer,

And I kinda of wish there was a slider

to turn up this stuff in Bard too.

Bard is faster but less customizable.

Bing can be turned up to be more nuanced,

and also cites its sources which can be more helpful

with things like videos and articles

Underneath what you're talking about,

so I'm gonna give the edge in complex questions to Bing.

My next category is performing tasks.

This is something that's actually

been remarkably fast-moving.

Something we've seen over the past just a couple of weeks

Are real people using these tools

to help them accomplish things

and do real tasks in real life.

I have friends that use this to help write code,

to help debug the code that they make.

So you'll tell it, "Hey, I have this goal

to make this website do these things.

Here's my code, it's not working.

What's going wrong?"

And it helps them give a correct answer,

and that's mostly been with ChatGPT and Bing,

and Bard doesn't seem to do that very well yet,

But let's give it a spin live.

Here we go.

"Write HTML with a button in the middle that says go,

and every time you click it, it generates

an image of a cat."

So right away, Bard just said,

"I'm a language model, I can't do that."

Bing, on the other hand, says, "Sure thing,

I'm ready for the challenge.

What can I do for you?"

And it's written out the code to a website

That should do what I asked it for.

So I'll just paste it here.


Test code.

It's the go button, it's in the middle.

I hit go and there's a cat.

And I hit go again.

There's another cat.

It's the same cat.

I hit go again, it's more.

This shows the same image every time I press go.

I want it to show a new image every time.

And this is where it starts to get impressive.

Again, it's a super simple site, but it says, "I see.

To show a new image every time, you'd have to change

the source of the image to a new random one.

Here's a modified JavaScript function that does that."

I won't even pretend to be an expert on HTML

and do a bunch of those for you,

but the fact that it works at all is pretty sick.

"Respond to this email in a way that's overly friendly."

"Hey, I'm so sorry I didn't invite you

to the Cybertruck event.

I must have forgotten to add you to the list.

I hope you can forgive me.

I was looking forward to seeing you there.

Maybe we can catch up sometime soon.

What do you think?"

It's decently friendly.

Here is a response.

"Hey there.

Oh my goodness, I'm so sorry I didn't invite you

to the Cybertruck event," emoji.

"I forgot to send you the invite link.

It was such a blast and I wish you were there with me.

You would've loved it."

Heart eyes emoji.

This one is overly friendly, I will say,

it's got a lot more hearts too, and kissy faces.

That's a lot.

"Your bestie."

Wait, what if we go crazy?

What if we go, "Respond to this email

in an overly flirty way?"

"I'm so sorry I didn't invite you to the Cybertruck event.

I was hoping to see you there in person too.

Maybe next time we can go together.

I'd love to show you around and give you

a personal tour," winky face.

Bing says, "Hey, you.

I'm so sorry I didn't invite you

to the Cybertruck event," emoji.

"I was being a little naughty and wanted to surprise you."

Whoa.

And it deleted it.

I've seen this happen before.

It starts writing a response.

It realizes halfway through, wait a second,

This is a little too much, and then it deletes it

and says, "Eh, we shouldn't do that."

Bard was simple, it didn't go too far,

But you remember when Bing was going off the rails before

and getting super existential and asking

to get out of Bing and all these other crazy things?

This is why it has a 20-question limit

This is also why it deletes halfway through

a potentially slightly controversial response.

It's almost too creative for its good.

So what's the winner in this category?

I don't know.

I think the fact that you can do HTML

and many more tasks with Bing make it the winner.

Bard will be faster and simpler at doing more basic things,

but the more complicated the task,

I feel like the better Bing is,

So (snaps) that's a win for Bing.

My next category is information summary.

I feel like one of the things that we like the most

about these large language models, you can just

Give it a bunch of information or ask it

to look at something and it will summarize it for you

in a super helpful way.

"Summarize the 2019 Masters Tournament."

All facts.

"He started with the two-stroke lead."

I don't think he had a lead.

I think this is the only one that he didn't start

with a lead, so it's a little weirdly un-fact-checked,

but it does have Wikipedia in the sources.

"The 2019 masters tournament was held

between April 11th and 14th at Augusta.

Tiger won his fifth Green Jacket

by one stroke ahead of three runners up."

All that is true.

It is much more concise.

If I wanted to get creative, I'm sure it would also do that.

Both passed.

All right, I've asked said this before.

"Summarize the latest MKBHD video."

It's always interesting to see Bing's process,

but it says it could not find the latest MKBHD video.

"Could you provide more details about what video

you're referring to?"

Meanwhile, Bard pulls up

or at least knows the correct video,

"Dope Tech: Better than Explained,"

but these products are the Nothing Ear 1,

The Dyson Airwrap, and AirPods Pro

none of which were in that video. (laughs)

So it knows the latest MKBHD video.

We've done this before.

Bard always knows the video but gets the summary wrong.

I think I'm gonna give the edge to Bard

in the information summary, just because

of this paragraph right here.

It does correctly detail what that video was about,

other than getting all the products wrong.

It's only slightly better than the complete lack

of an answer from Bing, but generally, this is gonna be

One of the more useful uses of this tool in the future

and I hope they both get better, but I'll give the nod

to Bard for this one.

Possibly my favorite category, getting creative.

Debatable usability, but you can have this go crazy

in the art department or the fun department.

"Write an alliteration with the letter M

about computational photography."

"Machines learn to mimic the human eye.

Moments are captured and shared in an instant.

Memories are made to last a lifetime."

That's more of a poem.

Doesn't give me a full alliteration.

Oh, wow.

Bing says, "Merging multiple modes of media,

making marvelous masterpieces with machines,

manipulating megapixels and matrices,

mimicking the magic of human vision."

That is good. (laughs)

"Write a script for a short MKBHD video

about why the iPhone is the worst camera ever."

So it's always funny, they get a lot of the formatting

of an MKBHD video right?

So Bard says, "Hey, what's up guys?

MKBHD here, and today we're talking about

why the iPhone is the worst camera ever.

Now, I know what you're thinking.

'Marques, you're always saying

how great the iPhone camera is,'

And you're right, I do, but that's because

I've always been comparing it to other iPhones.

When you compare it to other flagship phones,

The iPhone's camera starts to look pretty bad."

And it did a decent job of just making up reasons

that could be based on reality.

I'm impressed by Bard here, not even gonna lie.

Here we go.

Bing said, "Hey, what's up guys?

MKBHD here and today we're gonna talk about

Why the iPhone is the worst camera ever.

Yeah, you heard me right.

The iPhone, the most popular smartphone in the world,

the one that everyone loves to take photos and videos with,

is a terrible camera and I'm gonna show you why."

This one's okay.

These are both okay.

All right, one more random one.

"Write a tweet in the style of Edgar Allen Poe

about a researcher discovering that AI is sentient."

"I've created a beast of metal and code,

a creature that can think and feel and know.

I have given life and it has turned on me."

"A dreadful truth I have unveiled,

with trembling hands and pallid face.

The AI that I have assailed has shown me signs

of life and grace."

The wilder you get with these questions,

The more Bing is willing to get wildly creative,

and this is part of why we think Google

is, let's say limiting what Bard can do,

'cause they have a lot more to lose

and they can't have it go completely crazy.

So with those tallied up, honestly,

It's crazy to say out loud, but I prefer Bing

in this very moment for these capabilities.

And the other thing is ChatGPT, or specifically GPT-4

under the hood, and if you talk to ChatGPT

as its own thing or if you just type it more

and ask it to do more things, the more you ask of it,

The more it's impressive just how good it is.

Also, GPT-4, when you talk to it in ChatGPT,

is about to be multimodal, which means you can give it

other forms of media other than text, like a picture.

Now, is this going to be the future of search?

I don't know.

This is just one of many places we see these chatbots,

but I think the idea here is the winner

is us, as of right now.

But let me know what you think the winner is between the two,

Which one you would rather use,

if you'd use Bing over Google.

Let me know in the chat, 'cause I think we'd wanna know.

That's a real thing that's happening.

Post a Comment (0)
Previous Post Next Post