Welcome to another episode

of Cloud Unplugged.

We have some big stories today,

three stories and then a secret story.

Saudi Arabia is going to be

the new AI future.

Carbon emissions,

they're going to be

detected or are being

detected at the moment from

space along with nitrous oxide.

But does anyone care?

by now.

NVIDIA robots are coming for

all of our jobs as well.

So watch out for that.

And a random story around,

we're about to find out

what dolphins think of humans.

So might as well crack on, Lewis.

How has your week been

before we get into the stories?

Good?

My week has been quite good.

You've tried to take me up

to sort of slightly more

Loftier thinking, which has been good,

but disruptive on the coding away.

I had a good weekend, though, with no tech,

with use of saws and drills

and doing all sorts of things,

DIY styling for the garden.

So no vibe programming with

the sun this week.

So it's a proper tech-free week.

Less hands-on tech, for sure.

Well, actually, including me in that,

it's been derailing you

from doing any tech work.

It actually officially has

been a tech-free week, basically.

Kind of has, yeah.

Oh, well, never mind.

We have the news around Saudi, right?

Saudi Arabia,

and there was a summit that's

happened and a big

positioning for them to

become essentially the new AI hub.

So obviously they are

wanting to get Saudi Arabia

on the map to be known for

technology and AI, not just oil.

And they have big plans on how to do that.

And there's been big

announcements and big

amounts of money being thrown around from,

well, Nvidia, Amazon, all related to,

obviously, the sovereign wealth fund.

that's out there.

So what are your thoughts?

I have my thoughts on this

and opinions on this.

What are your opinions of

this big announcement for

Saudi Arabia to become the AI hub?

Yeah, it's kind of mixed messages.

I think from a hundred feet up,

there's a bit of context

around how these announcements came.

It's geopolitical and

with a long term direction from Saudi.

So the geopolitical aspect is, you know,

the Trump government

meeting in Riyadh with

crown prince and and working

through this uh bunch of

announcements for closer

collaboration not just in

ai um across energy sector

I'll deal on with all of

his bits and all of his

things he can do in saudi

deregulate that get in there um and then

there's a whole bunch of

reciprocal announcements

about the sovereign wealth

fund investing in US data centers.

So it's a bit wider.

But then taking it back to your question,

sorry,

I couldn't resist putting my

hearings on it.

And it seems perfect, you know,

AI investment in a post-oil economy.

You know, it's a great general direction.

And we can dig into some of

the more specific bits.

But I think there's maybe a

bit of hypocrisy in this space.

Hypocrisy how?

Sorry, I don't understand why.

Why do you think there's

hypocrisy on them being an AI hub?

So...

Not really on the AI hub itself,

but the proposed benefits of an AI hub.

So the benefits locally

would be developing AI

foundation models that understand

Arabic.

Sorry, how do you say it?

How do you say the language name?

Arabic.

And Arabic foundation AI models.

Obviously,

if they understand the language

and the LLMs become fluent

or are trained to be fluent,

then there's benefits for local jobs,

local economy,

and the beneficiaries can

be the population of Saudi Arabia.

and obviously post oil um

implies you know a green

economy and you know that's

the twenty thirty

initiative um that the uh

the saudi vision um project

has so it's all positive

but the hypocrisy is the

track record you know the

funding for all this uh and

the actual main customers

so the funding first is the pif

which has got a terrible

human rights record,

as has the government.

Amnesty International and

the Human Rights Watch both

have reports that have dug

into various crackdowns on

migrant workers,

on indentured workers and, you know,

quite

quite stark punishment and

authoritarian practices.

So I guess the hypocrisy is

more how it's sold

internally as benefits to

the local populace versus the actual,

what that means to

people that live in Saudi, you know,

small tribes that are moved

to make room for new data centres,

new building projects across the kingdom,

really.

And, you know,

the benefits to local jobs

and the economy might be

offset slightly by the mass

surveillance and crackdowns

on LGBT rights, et cetera.

So... Yeah, I'm not a fay with...

the political details of Saudi Arabia.

I do know, though,

that I think in certain places in Dubai,

obviously,

they're quite good for people

that are local in the area,

especially with all the

buildings and the hotels

and the workers and

accommodation and tax

incentives and things like that.

However, yeah, I guess

I don't really know enough

about the ins and outs of

the political landscape.

But I do know their fund is like massive,

three point seven trillion

pounds worth in the fund,

which is like the PIF funding,

which we kind of spoke about.

Then there's also they've created a VC.

I don't know if the VC is

created by the same fund or

an arms thing.

It's fully owned by the fund.

It is, isn't it?

Oh, wait a minute.

They're different funds.

You're talking about Wilde.

so there's wild which is a

bc which is which is

venture capital that's just

pure startups founders

doing ai they've got a mil

to invest there but also

are looking to invest in

san francisco too so I

guess to fund things to

then consume which makes

sense right go and do the

funding in the states

and then also get them to

use um the cloud we spoke

last time in the podcast

about them changing the

sovereignty rights of that

data so that's like another

move which is like actually

don't worry about the data

sovereignty laws you're

still on the data ready for

this big announcement on

the data center investments

that have come in there's nvidia

that's going to be

installing obviously their

chips and the hardware for

the data center for

omniverse which will help

them I guess for the

manufacturing side of stuff

and you know modeling um

you know twin modeling you

know a replica of all of

these things to obviously

work things through using

the video and amazon then

is kind of upskilling and

putting infrastructure in as well um

A new region, a new AI hub.

A new AI hub, yeah.

They're going to build the

entire infrastructure, yes.

Exactly.

So they're really going for it.

They're laying it all out, right?

So all the path is being laid.

You know, to be fair,

they are actually thinking

about all the right things

because they're like, okay,

who's going to host over

here if we own the data?

People are obviously going to resist that.

We obviously need to fix that.

That's going to be a block.

They're certainly very, very serious.

Yeah, very serious.

I think there's five hundred

billion or something

ridiculous overall in this

space on this one.

um set of ai initiatives um

spread across numerous

initiatives like the um aws

the um but the there's a

new separate company called

humane yeah which is

Maybe an ironic name,

given the things I started with.

But anyway,

there's a company called Humane,

which is fully owned by the

PIF and has strategic

initiatives with both NVIDIA,

a five billion AI project,

and DataVault's NEOM data center.

So big investments locally in Saudi,

but also reciprocal AMD and

Humane building data

centers up to the tune of

ten billion in the US.

So is Humane concentrating on just Saudi?

No, it's an AI specialist.

Yeah, I mean, they're definitely,

definitely

know very serious on the

money they're always

furious on the money but

there are you know there's

a context of uh massive

crazy projects um that the

saudi government have been

announcing for the last you

know over the last five to

ten years and the line

probably being one of the

most and project neo um which is um

just a tech centric,

complete rebuild of the

economy to not be dependent on oil.

But I mean, the line is meant to be this.

They started with a hundred kilometre long

building for a new city

that's entirely green,

and then it became one kilometre.

Now it's not going to be a kilometre,

so it's being scaled back further.

So maybe it's about to disappear.

I have no idea, but they're building it,

and they're moving tribes

away from it and getting on with it.

So they are definitely

serious on the money and

the intent initially, where it will go.

It's early days.

Yeah,

I think it's quite... Romania's brand

new.

In some ways, quite smart,

because economically,

if you think of their trajectory,

it's very hotel-based.

There's a lot of investment in businesses.

They tax incentivise you to be over there,

right?

So they're making loads

of... Are you talking about

Dubai or Saudi Arabia?

Totally different places.

Well, but even Saudi Arabia is the same,

isn't it?

So they're building the one

kilometre thing.

That was the whole point.

So they're going to do loads

of manufacturing.

That's all Saudi Arabia, yeah.

Yeah,

so you do loads of manufacturing and

trying to get people over there,

trying to do tax incentive,

trying to do levying on the

data sovereignty so that

they are maneuvering in to

put their wealth to work to

try and get investment in.

So the aim is, well,

if you're going to attract people over,

you need to do investment.

and how do we get all the

investment over here well

we need to invest in things

to get the investment over

here so we need to

obviously be investing in

companies that are going to

be ai startups to get the

investment over there and

start to attract them we

need the data centers in

here so that I think

obviously trying to

incentivize as much as

possible to try and get money in

Saudi Arabia.

The template of Dubai in the

region for how you invest

and what you can invest in.

And obviously the line and

projects like it,

they've got huge ski

resorts and mountain regions.

So they obviously want to

get more on the

international stage generally.

And then all these AI projects.

related announcements to

give people jobs in a new, interesting,

exciting... Well, to make money, I guess.

Economically,

it's to increase the wealth

and de-risk from other

economies that maybe don't

have as much of a future.

And obviously being influenced maybe by

you know,

others and Trump and Trump coin

and Melania coin and all

the meme coins and all the

other things kind of going on, you know,

a lot of other things happening.

One of the other stories is,

which isn't that new

because we're talking about before,

but emissions and tracking

the emissions from space.

So nitrous oxide, carbon dioxide,

methane and the EMAP or NMAP

guess satellite that is like

detecting through light um

and the reflection back off

the earth and then

basically how it gets

absorbed so they can kind

of check spectral analysis

spectrum analysis exactly um

But, you know,

they've been talking about this.

It isn't necessarily that new.

There's been advancements in

this space where, you know,

the range of which they can

detect is like very detailed now.

So they get very,

very precise on the emissions.

Essentially start environmental mapping.

But fundamentally,

does anyone really care about the data?

So the data comes in.

You know, they can see, yeah,

this pretty bad or maybe pretty good,

but the data needs to inform policy.

And, you know, the US has been involved.

China's been involved.

They've been doing loads of

like sharing and workshops

and all of these things around this.

And obviously NASA's

involved in additional

satellite capabilities as

well that do very similar things.

However, out of all the countries,

which I think it's like a

hundred and twenty countries,

probably the main ones are

going to be China and the US.

And the US really,

I don't think necessarily

cares because they rolled

back a load of their policies,

something like a hundred and twenty.

Doge again,

your best friend Elon went and

dismantled a lot of the

environmental departments,

got rid of a load of people.

So what are your thoughts?

on this technology,

actually its usage globally in the end.

I guess it's effectiveness

on the world in the end

from a technological perspective,

but actually more interesting.

Yeah, from a purely technical perspective,

perspective,

a technological perspective on

environmental monitoring from space.

This has just been something

that's been increasing over

the decades since the first satellites.

I mean, you know,

environmental monitoring

from space using spectroscopy is

started in the sixties, seventies,

eighties, you know,

it's just with the

shrinking electronics and

more launch capabilities,

there's a lot more capability.

And indeed, I think the actual study,

one of the studies that

were cited in this NMAP

reports were talking about

the volume of data and how

wide it is and how real time it is.

It's completely current and

can be used to enforce

local policies by

governments or by

scientists doing climate studies.

So technologically,

I think it's just a

continued increase in capabilities.

Basically, we're almost Star Trek now.

Scan the planet for life forms.

Beep.

Blah, blah, blah.

Job done.

So we're at that sort of

increased capability stage.

It's exciting from a scientific and... Oh,

I think it's really, really positive,

but from a...

political standpoint and

also just that but also an

information age where

people just basically you

know rebuke anything that's

said anyway oh yeah those

stats are nonsense anyway

you know I've got another

data source where

apparently it's these

emissions and actually

you're talking nonsense

it's bullshit and actually

these satellites actually

NMAP satellites that's

absolute bollocks anyway

you know what happens and

then it just goes on and

everyone's like yeah let's do nothing

uh I think there's there's

there's some there's some

optics certainly

geopolitically I mean us

seems almost schizophrenic

and well maybe their their

great leader um is slightly you know

And it's definitely phrased

in the one hand,

we're going to do all these

amazing things,

the best that's ever been.

And then on the other hand,

completely dismantle all

the things that we're doing.

You know, they don't necessarily add up.

On the one hand, you say this,

and on the other hand, you do that.

So I think there's a moment

of pause with complete dismantlement.

Quite a lot of the NASA

funding has been fantastic, thinking,

well,

let's remove some of the crazy

bureaucracies on the SLS system,

but let's remove some of

the climate modeling and

science programs.

and a large part of NASA's

budget was towards with

climate monitoring and

satellites in this

monitoring space so that's

obviously very worrying but

I think kind of with a

pinch of salt because

scientists doing science as

long as they're funded I

suppose is the issue you

know they are doing the

right science and the

scientific capability is

moving forward but

Yeah,

I guess your point is if they're not

funded, then they can't have a job.

So although lots of people

are interested in doing the

right thing and the capabilities there,

who's going to fund them?

Yeah, I think there's a few countries.

and I think a lot of this

comes from universities

anyway so even universities

are getting attacked in the

states right so I think

there's quite a lot of

pincering going on maybe

for different reasons and

it's just you know it's not

like some strategy it's

just a bit chaotic and just

so happens to have you know

there's a lot of um

symbiotic relationships

between everything like how

things work through the

investment in science

inside the university is

invested you know for

certain reasons that

progresses things you know

globally you know from

advancements versus yeah

the policies and people

reporting the policies but

one thing I will say is

that you don't hear about the actual data

So I couldn't tell you.

I mean,

I probably could have done more research.

That's probably me.

But I couldn't tell you

actually statistically what

it actually means globally

from these emissions and

the predictions that

they've come up with.

And I probably wouldn't

spend the time personally, obviously,

with everything else competing,

as most people wouldn't,

unless you're really interested in it.

digging into the research

and all of the data,

somebody needs to be

communicating and

simplifying the information,

and that isn't really happening.

Well, it might be, but I'm not privy.

Yeah, it's not clearly,

and I bet I could ask quite

a lot of people and ask them,

and I don't think they'd know,

which means it's not mainstream.

There have been...

Yeah, no, you're absolutely right.

But I guess what actually

affects outcomes is the

local policies and politics

around that and the optics.

And there have actually been, ironically,

there's been some really

foul noises coming out of

the UK government,

but there's also been some

positive things about

environmental monitoring

and testing of policy to

see that the outcomes are actually met.

I think China

as you cited,

have domestic policies around

reducing emissions.

And using this technology

means there's actual real

measurable outcomes on

pollution and public health,

which I think are really positive things.

And I guess we've just got

to see how regimes change

and which are the good ones,

which are actually

benefiting their people.

And, you know, I don't know,

if democracy holds,

then maybe things can get better.

I can't really say China and democracy.

No, I'm saying in the state,

if democracy holds,

then maybe things can get

better and policies can align.

But you're right, it's weird.

It's not one type, because obviously China,

not being a democracy,

seems to be absolutely

crushing it in the environmental states.

Yeah,

I think they've positioned themselves.

Knowing how clean the air is.

Yeah,

they've positioned themselves on

their big plan,

because they've got more

like ten-year plans, don't they,

in China?

And as it is a big... You

need to have a

ten-plus-year plan for climate change.

So in some ways,

democracies will change

parties all the time for a

problem that is...

not a three or four year endeavour.

Obviously,

it's problematic in itself with

a democracy.

It's like,

how does anyone get a chance to

fix something that's going

to take more than four years?

And then you could

potentially keep switching

the party every four years

and never really address

the problem because you

keep changing the approach constantly.

So I could argue democracies

aren't really suitable for

big global long term challenges.

Who knows?

Anyway, let's go to China, Lewis, quick.

Well, you were in China not that long ago.

I was.

It was very green, very clean.

Lots of electric cars, electric scooters,

electric motorbikes, electric buses.

It was all very... You could smell it.

So maybe their policies are

already working.

Yeah.

What is your... Moving on then,

away from that, the...

Groot, Groot N-One,

which is to explain for people.

So NVIDIA have launched an AI model,

like a foundation model

that's trained on human, I guess,

behaviour,

generally how they walk and move.

And

They've released it and it's open source.

So, you know,

the first open source human movements,

basically synthetic data trained model.

And they've based it on...

Have you ever read Thinking

Fast and Slow by Daniel Kahneman?

No,

I think you've mentioned it to me before.

Yeah, so there's a book.

I think it's only...

but you might be able to

listen to it or audible or something,

but he did a great book on

basically the two systems,

humans have the system one

where you don't really have

to think um because it's

like quick for your brain

so lizard lizard style

innate behaviors um things

that you can say like one

plus one you'll know the

answers to you don't think

about it right um learned

or trained just trained

yeah so and then there's

things where you'd be like

twenty five times three

thousand six hundred and

twenty four right you'll be

like oh my god I don't

no,

I've got to really think and engage a

different system,

which is now actually the

pragmatic thinking system

to calculate things and

work out the answer.

And so his book is based on

a system one and system two.

And it then also,

which just not to go on too

much about the book,

but it also looks at the

psychology because that's

kind of the premise of the book,

which is how your behaviors

around each system have

different trust models.

So you'll trust more your innate.

So things you can see.

And they did an experiment on, I believe,

a...

basketball match where the

ball's been passed and they

made people watch and

they'd be like right the

people in the black shirts

count how many times the

basketball gets passed so

then you're focusing in

like counting and then

somebody in a gorilla

outfit runs across the

pitch right and because

they're concentrating they don't notice

And then they interviewed them saying,

did you see the gorilla?

And obviously,

because they have such trust

in system one,

they thought the person was lying.

They wouldn't believe that

they couldn't acknowledge a

gorilla running across

you know the pitch and so

you end up basically

refuting facts because of

the trust system built on

things that are just innate

right so so how does this

relate to uh just just just

just out of interest I did

I did say unrelated yeah

yeah I did caveat that

um but anyway it's been

trained they have a system

one and a system two in the

training on the synthetic

data and actually they have

done a very similar thing

which is language and

reasoning and understanding

so they've got system two

they've got system one

which is obviously reflex

and so you know if you're

trying to work out a path

and a route right so

through a complicated

forest then you might have

to think about okay what's

the direction into going

versus if I'm just going up step by step

you know it's a little bit

more in the system one or

one foot in front of the

other so they're very

different interactions that

your brain and they've done

them in a very similar way

where they've obviously

trained the model so it's

actually based on system

one system two for the new

rise up of these robots

these humanoid robots that

are basically coming for everyone's

I think it's fascinating,

the NVIDIA announcement

over the open source in their model,

but also how it was trained,

how foundational,

how it uses synthetic data

based on simulating things

in a sped up universe.

Because to do physical

training of real robots is

obviously very expensive.

But to do simulated training

of simulated robots in simulated reality,

you can run it many times

and you can repeat the

experiments very fast.

So you can get a lot more

innate system one instant behavior,

which is required in

robotics for immediacy,

understanding your

environment and catching a

ball or balancing or

dancing and not moving robotically.

means system one you know

absolutely and then system

two for planning and

working things out but

system one is absolutely

required for balance and

reasoning and dancing and

just moving naturally and

being able to yeah things

you don't think about at

all yeah you just your body

just does it I mean and the

fan fascinating thing is that the uh

It's not just NVIDIA's model.

It's great that they want to

enter this space.

But actually,

across the humanoid robot space, I mean,

the Boston Dynamics latest robot,

they've showed them doing

all sorts of parkour and things.

But their latest humanoid robot using

synthetic data training

shows a very natural movement.

And just a couple of days ago,

the latest Tesla robot dancing,

showing it's moving very,

very fast and dynamically.

And all of the Chinese robot

manufacturers have shown

very similar things based

on entirely synthetic training data.

So they've been able to

completely train them

about the real world using

models that closely

resemble the real control

systems of a robot,

and then take them to the

real world and prove that

they can really do those things.

So it's been a long arc of

getting the data,

getting the simulations good enough,

because actually getting

the pipeline for training

is a bigger job than we've

got robot and got models,

but they're crap,

and training them with humans,

teleoperation or whatever,

is going to be extortionately slow,

excruciatingly slow.

So training with synthetic

data in models like this,

I guess this is about

democratizing training of

robots using... Well,

I think if we strip it back

on what it means in a video,

first you just said so,

I guess number one,

they've got the Omniverse

platform that they can

leverage for these

simulations to train on in

a simulated world.

So they've already got an advantage there.

And then two, the market,

the potential market for, you know,

humanoid robots is insanely massive,

right?

That market potential.

Even if they're just walking dogs,

it's massive.

But if they can move

naturally in the space,

and even if they're just

carrying your luggage as a porter,

they don't need to have too

much system too.

They don't need to be crazy high brain.

A walking dog.

Does that mean if you could

get a robotic dog that was

just like a dog, but without all of the

Oh, no, no, no, no, no.

I was saying a humanoid

robot walking a dog.

But if you want robots walking robots,

you could do that.

Are you talking about a

human walking a dog?

No,

a human robot at one of its earliest

jobs could be walking a dog.

Oh, I see.

You didn't mean a robotic dog.

No.

Because they are robotic dogs.

They are quite famously lots

of robotic dogs.

Right, okay.

Let's just close that loop

and get them all done.

But worryingly, would I trust, I mean,

you've mentioned Tesla,

would I trust a

megalomaniac individual who

has insane wealth and power

with humanoid robotic dogs?

system I I mean you know it

does make them god what was

that film with the the

tesla actually get all

taken over and yeah um it

just feels like starts off

as a film or a black mirror

episode and starts to

become reality doesn't it a

lot of sci-fi in this space

that has already covered

the topic but yeah it feels

closer feels closer but I

think it's in I think it's going to be

Really good.

The only caveat for this, again,

geopolitically,

that I do have a bit of

beef with is the states

when they're saying, you know,

put the tariffs on, stop outsourcing,

manufacturing, you know,

stop basically building

things overseas for cheaper, you know,

disincentivize you

basically importing goods

and actually start to

manufacture in the states.

need to give people jobs the

working class their jobs

back but then if it's all

robotics anyway there is no

working class job so really

all you're doing is

incentivizing manufacturing

but probably with a plan to

be more robotic based.

And I'm sure Elon's in the air saying,

look,

we can manufacture all this much

cheaper anyway,

if we could start having

all these robotics.

But under the guise of

pretending that it's for

the people inside the

States to get jobs back,

which people don't want to

do manufacturing jobs anyway.

Culturally, it's probably moved on a bit,

but

Yeah, it's a double edged sword.

What can provide amazing

sociological benefits for caring, for,

you know,

help of in health care settings

and for the disadvantage

and to remove manual labour

and dangerous labour and start to.

give people a focus on

working directly with other

human beings using the

skills system too and

things where you reason

about people's wellbeing

and care and like actual companionship.

But do you not think that's insane?

That's insane to me.

It's like you've got a

population that's exponentially grown

and you're then saying do

you know what what we don't

have is people to look

after people so what we

need right now even though

we've got like an insane

population like crimes from

like from previous hundreds

of years we need robots

that's what we need we need

robots that can look after

people it's like honestly

we don't we actually don't

we actually have enough of a population

And there should be community incentives.

I think you're right.

We don't need robots.

If there's one benefit,

it's refocus what people should be doing.

And we should be spending

more time with each other

and families and the

elderly and have more of a

cohesive society and get

all the drudgery done for us.

And if that means a bit of robots,

maybe we do.

Don't know.

Complex.

Yeah, I'm not...

or robots necessarily,

but I do feel like it's a

slippery downward spiral.

I mean, I mentioned the film iRobot.

What could go wrong?

Not even iRobot.

I just think, though, I mean...

You've got DoorDash and

Deliveroo and Uber Eats and

you've got bikes that are electric.

You don't even have to pedal anymore.

It's like at what point you're not cooking,

you're not walking, you're not cycling.

AI is doing your thinking for you.

I mean, what are you even doing?

What are you even actually doing?

Well, in the end,

the story Hitchhiker's

Guide to the Galaxy,

I think the third episode

was about dolphins being

the clever ones and us being the thickos.

So maybe we are.

That's a segue.

Well, we'll find out.

That was a really good segue.

Wow.

Honestly, I'm actually quite impressed.

That was a really good segue.

System two thinking there.

That was pure system two thinking.

No robot would do that.

No, no.

It could never do that.

That was too advanced.

Yeah, okay.

Well, Segway, you've stolen the show there,

so you should set your Segway then.

I can.

Yeah.

I mean, Dolphin Gemma is the story.

And who's she?

Gemma.

She's the sister of Gemini.

um who's who's like another

ai actually you you phrase

this would you you've got

the the the initial thing

and then I'll I'll leap on

in yeah so they google uh I

mean fine I'll I mean you

did the segway it feels

like I'm cheating but um

Google partnered with some

marine biologists for the Google Gemma,

which is basically an AI

model trained on dolphin, I guess,

can you say speech?

It feels like something you,

I don't know why,

it feels very human in language,

but I suppose...

I mean,

just talking to you don't understand.

Honestly,

I did because Google Gemma's on my phone.

I literally did understand

everything you just said.

Honestly, Lewis, I totally agree.

I totally agree with

everything you just said.

um yeah so they're basically

being trained on their I

guess vocalizations the

pitches and the noises and

to then try and turn it

into a language model and

then try and decipher I

guess the communication

that's happening to um

interpret dolphin um yeah

so yeah I mean it's

absolutely fascinating I

mean the model insane yeah

the dolphin gemma model I read um

is a tiny model.

And it's like an LLM,

but trained on acoustic

sound as the token space.

It's not trained on ASCII first, I guess,

on character space.

It's trained on an audio representation.

They've had to work out how

to get the salient bits of

the audio into their model.

And it's predicting the next generation

audio noise that would go with them.

So they've done decades of

research in this space and

there's an enormous amount of data.

But the problem is, well,

how do you distill that

data down into something

actionable where you can

then use it in your research?

So what they've done is made

this Dolphin Gemma model,

which is relatively small

because the foundation

isn't the whole internet

because the internet isn't

written in dolphin, weirdly.

It's just the researchers' recordings,

and they've been able to distill that,

given the correlations that

they've really been

hard-won over many decades

of what's a name and what's

a more nuanced meaning verb,

or which bits of vocabulary

can we discern from decades of research.

And then, once we've got that,

can we predict anything from it?

And they've got this model,

which is only a few hundred

million parameters,

which is tiny in relation

to the eight billion tiny

models that will run on

your desktop and the thirty,

fifty to a hundred billion

parameters models that are

used in language these days.

So these ones can run on a phone,

which is the interesting thing.

But I mean.

two things why the fuck

would you call it dolphin

jenna honestly I mean that

that is a bit crazy

honestly you should have

called it a better name but

the second one till I tell you why

Oh, go on then.

Oh, yeah, yeah.

Uh-huh, uh-huh.

Right.

Ah, I see.

Okay, that does make sense.

The second thing is I don't

think they've got much to talk about.

Like, I generally don't think dolphins...

really have that much going on,

generally speaking,

to have an extensive vocabulary anyway.

Would it not be fascinating

as a scientist working with

a marine scientist

biology, in that setting of that species,

dolphins,

working with this data to

actually make breakthroughs

where they can respond and

get reactions and start to

learn or decipher or have

any type of empathy at all with another

intelligent to a degree for

sure I mean the flippers

are a little bit of a shame

we've got these opposable

thumbs the flippers thing

mean they haven't

progressed their society in

lots of ways but as you

were just saying do we

really need robots should

we be spending more time

with each other maybe the

dolphins have got it all sorted

Yeah,

but maybe we're about to interfere in

something we shouldn't

really be interfering in.

Maybe we'll learn something.

Oh, maybe we will learn something, yeah.

But it does feel like one of

these things that could be... I mean,

I think it's amazing

because it's just the idea

of being able to actually

interpret Dolphin and also

potentially speak Dolphin.

Whether people, I mean,

obviously you can speak through it,

Dolphin.

Hold your phone up again and I'll...

demonstrate I mean I do feel

like I I believe I but I

mean having interpreted

everything you've done

already lewis I do believe

but um yeah it just reminds

me of those things where

because you people already

know quite a lot I imagine

the marine biologists about

dolphins because they've

researched them for so long

it just becomes probably

like quite obvious that

those were the things that

they were kind of saying a

bit like a garmin watch

when you like people wake

up and they're like

you know,

have to look at their watch to

see if they're tired and

got a good sleep or like, you know,

oh God,

I wonder how many steps I've done.

It's like, well,

you either walked a lot or you didn't,

I mean,

one way or another,

you should know these things.

You don't need the watch to

inform the decisions you've already made.

So I do find those things quite funny.

Obviously, I am a bit pro-watch.

I kind of get it anyway.

I'm just being facetious.

But it did remind me that

you just find out that, yeah, yeah, yeah,

I suppose that was what we thought.

Yeah,

they were the things I thought they

would be saying based on

all of the research that

we've been doing.

But yeah, it's super, super interesting.

I do like that idea of speaking.

We're in a very interesting

time where we're speaking with machines,

maybe dolphins,

and there's more than one intelligence,

or at least on the spectrum,

thing going on.

And there never has been

before in human civilization.

Yeah, it is massive progress.

Yeah, absolutely.

Well, they are the stories for this week.

I've got to go to Saudi.

Are you coming along with me, Lewis?

We need to get on our flight

and do some modelling on the new AI hub.

I need to get you translated.

Maybe teach you a bit of

whale whilst we're over there.

Could do that.

All right, let's do it.

All right, well,

nice being to everybody and

I shall see everyone soon.

Adios.

Adios.

Adios.