there's so much to talk about in

technology now the title is debating

technology I don't think there's a

debate technology yes or no um but in

covering Silicon Valley for 25 years I

often hear you know technology can be

used for good or bad which is inherently

true but sometimes that's used to say

especially by the makers of the

technology well it's going to be used

for good or bad hopefully the good

outweighs the bad and to me that

neglects our responsibility to push and

steer and limit the tech techology so it

is used for good um but we're going to

talk about this is a moment of great

excitement especially with artificial

intelligence Robotics and all these

Technologies um but it's also a moment

of great concern a lot of people have

legitimate fears about what this change

will bring uh that's enough from me I'm

excited to be joined by David Newman

head of the MIT media lab and Yan Lon

who leads AI uh research and other

activities at meta um Dava maybe to

start with you I mean you have such a

broad background in technology from

obviously your experience in space where

is your head these days what are the

problems uh and areas that you think

need our attention and where are you

wrestling your brain around thank you

everyone good morning pleasure to to be

with

you so where's my brain typically an

outer space you know thinking about

becoming an uh you know uh inter species

will we find life elsewhere it's not

option b so where my head really is and

thinking about technology and the

disruption

that we feel and that much more orders

of magnitude more disruption that's

coming so maybe I paint the picture it

really is I think a a you know

technology super cycle now convergence

of probably three Technologies at once

you know the Industrial Revolution that

was it was okay when we put one

technology at a time geni took me 30

seconds before getting into this AI it's

it's coming large langu it's but it

still it's an infancy at the MIT media

lab we've been working on AI for 50

years so now that it's uh common in

everyone's hands a co-pilot I'm sure

we're going to debate that and talk a

lot about it with my esteemed colleague

and and expert developing that we're

doing a lot most important thing I want

to emphasize just in the introductory

about Ai and gen AIS we design for

humans human centered human flourishing

at the media lab so is it trusted is it

responsible that's that's the premise

actually we don't do it if it's not but

hold on to your seats you know everyone

um rocket launch is coming soon soon I

think we'll all be talking about gen bio

if you're not already not just synthetic

bio but generative bio we don't bio

biology is is is organic so when AI

morphs into to gen bio it's no longer a

large language model but we're working

on actually media lab um you know large

nature models now you're ingesting

biology and genetics and biological wrap

that all around into sensors internet of

things we're pretty famous for I call it

now internet of all things because I

have iot for the oceans to monitor all

biodiversity for the land climate air

atmosphere you might think of and from

space more than half of all of our

climate variables are now measured from

space so that so hopefully that kind of

technological Whirlwind I don't know

what else to call it you know coming

with geni Gen bio sensors measuring

everything to finish up I put humans and

human centered design right in the

middle and asking The Upfront questions

is it intentional for human flourishing

and all living things flourishing if the

answer to that is no with our algorithms

then I don't think we should be doing it

and Yan that's a that's a good point to

turn to you how do we make sure uh the

AI we can build is the AI we want how

are you trying to focus your work and

the development at meta to make sure

that we get an AI that works for

Humanity

um there's two two answers to this the

first thing is you try to make it work

well and

reliably uh and uh the the the flavor of

generative AI or AI that we have at the

moment um is not quite where we want it

to be in terms of is very useful we we

should push it we're pushing it um

trying to make it more reliable trying

to um um make it applicable to kind of a

wide area of a wide range of of of areas

um but it's not where we want it to be

and it's not very controllable um for

various reasons so I think what's going

to happen is that um within the next

three to five years we're going to see

the emergence of a new brand or Paradigm

for for AI uh architectures if you want

um which um may not have the the the

limitations of current AI systems um so

what are limitations of current systems

there are four things that are essential

to inell behavior that they really don't

do very well one is understanding the

physical world second one is having

persistent memory and third and fourth

are being capable of uh reasoning and

complex planning and llms really are not

capable of any of this um there is a

little bit of an attempt to kind of Bolt

some wordss on them to kind of get them

to do a little bit of this but uh but

ultimately this will have to be done in

a different manner um so there's going

to be a another revolution of AI over

the next few years and we may have to

change the name of it because it's

probably not going to be generative in

the sense that we understand it today um

so that's that's a first point some

people have called this in different uh

uh names um so technology we have today

um large language models deals very well

with the discrete world and language is

discret

um I don't want to upset stepen who is

St Pinker is in the room here but to

some some extent language is

simple

U uh much simpler than understanding the

real world which is why we have ai

systems that uh can pass the bar exam or

solve equations and things like that do

pretty amazing things uh but we don't

have robots that can do what a cat can

do um the understanding of the physical

world of a cat is way Superior to

everything we can do with with with AI

um so that tells you the physical world

is just way more complicated than than

human language um and it's because why

is language simple it's because it's

discrete objects and the same with DNA

and proteins right is discrete so so the

application of those generative methods

to this kind of data has been incredibly

successful because it's easy to make

predictions in a discrete world you can

never predict what word will come after

a particular text but you can produce a

probability distribution of all possible

words in the dictionary and is only a

finite number of them if you want to

apply the same principle to

understanding the physical world you

will have to train a system to predict

videos for example right show a video to

the system and ask it to predict what's

going to happen next and that turns out

to be a completely intractable task um

so the the techniques that are used for

large language models do not apply to uh

video prediction uh so we have to use

new techniques which is what we're

working on at at at MAA but it may take

a few years before that that pens out um

so that's kind of the

um the first thing and and what that

pans out um it will open the door to

brand new class of of applications of AI

because we'll have systems that we'll be

able to uh reason and plan uh because

they they will have some mental model of

the world that current system we don't

have so they'll be able to predict the

consequences of their actions and then

plan a sequence of actions to arrive at

a particular

objective um and that may open the door

to U real agentic system is talking

agent Ki but nobody knows how to do it

and that's kind of one way to do it

properly and also to

robotics um so the coming decade may be

the decade of Robotics because that was

the first answer and the second answer

which is shorter um the way to make sure

that AI is uh applied properly is

to give the tools for people to build

diverse set of AI systems and assistants

um with which understand all the

languages in the world uh all the

cultures uh value systems Etc and that

can only be done through open source

platforms so um I'm I'm a big believer

in the idea that uh the way the AI

industry and and and ecosystem is

going um open source Foundation models

are are going to be dominant over

proprietary systems and they they're

going to basically be the substrate for

the entire industry they already are to

some extent uh and and they're going to

enable a really wide diversity of uh of

AI systems and I think it's crucially

important because within a few years you

and I both are wearing those smart

glasses right and you can talk to an ni

assistant

uh using those things and ask any

question but pretty soon uh we're going

to have more and more of those things

with displays in them and everything and

all of our digital diet would be

mediated by AI

assistance and so if we only have access

to three or four of those assistance

coming from you know a couple companies

on the west coast of the US or China is

not going to be good for cultural

diversity

democracy uh everything else you know we

need a very wide diversity of AI

assistance that can only happen with

open source which is what meta is um

being promoting as well well thank you

both I think that sets up up well for a

discussion and as a reminder this is a

town hall not a panel so we're going to

be bringing in both the audience here in

this room of incredible guests as well

as uh those on the live stream so the

first thing we did is we asked uh the

folks on the live stream uh there is a

slido you can join um also we asked how

would you like these emerging

Technologies uh to contribute to the

Future and we're not going to show all

the answers but here's a word cloud of

some of what uh folks have said so if we

just quick quickly look at

that well that's the question I'm not

quite sure how we get to the answer new

technology it's a blank slate all right

well I'm sure people you know talked a

lot about both what they're excited

about and what they're worried about um

you know I want you to get ready with

your questions in the room I'm sure

everyone has some but Yan I want to

follow up on the open source thing

because there's really a big debate I

mean as I said technology is not a

debate but the approaches we take and

certainly open source has all the

advantages that you mentioned um it

allows people all over the world to join

in only a few people are going to be

able to train one of these giant uh

models but a lot of people can make use

of them and can contribute at the same

time there's a real concern that taking

this powerful technology and giving it

to the world and saying basically meta

says here's our acceptable use policy

here's what you and can't do but to be

honest there's really no way of

enforcing that once it's out it's out

how do we make sure something is both

open source and

safe uh so what what we do at meta is

that when we distribute a model so by

the way we say open source but we know

technically those things are not really

open source because you know the code

the source code is available the weights

of the model are available for free and

you can use them for whatever you want

except with those restriction Clauses uh

you know don't use it for for for

dangerous things um so the the the way

we do this is that uh we we fune those

system um and Red Team them to to make

sure that at least to first order uh

they're not you know kind of spewing

complete nonsense and or or or toxic

cancers or or things like that but um

but there is a there's a limit to how

well that works and and those those

systems can be J broken you can uh do

what's called prompt injection so type

of prompt that will basically

take the system outside of the domain

where it's been fine-tuned and you know

you're going to get to its uh you know

uh kind of root uh uh things and then

that depends on uh what training data is

been pre-trained on uh which of course

is a combination of high quality data

and not so high quality data and Dava is

putting something like that into the

world I mean obviously there are

benefits to open sourcing that way MIT

is Pioneer in open source there's an MIT

license for open source I can't remember

it may even be the license that meta

uses um at the same time when you talk

about having this technology be human-

centered and putting humans and our

needs and concerns at the Forefront what

do you think needs to be done you talked

about synthetic biology and you know all

these things obviously there's a lot you

know there's a lot of neglected diseases

there's a lot of things we want to use

these new technologies for and we don't

want everyone just in their home

developing new microorganisms to run

around so what are your thoughts on how

we make this technology broadly

available but still safe yeah thanks

that's that's the question and seeing

what people are concerned about too AI

in space I agree with that we can talk

about the word cloud but so um you based

on open source platforms with but with

guardrails and and we have to be all

held accountable right now we can you

know ask the the audience as well you

know does AI work for you what I mean do

you trust it is it responsible is it

representative of you do you think it

has the training data that represents

you well let's ask the audience how many

of you feel that you think it's safe

secure and um you know you're going to

launch in and use it today you know

during this

debate anyone raise their hand well I I

think there's the answer and I think

it's not how many people would be open

to AI would love to use AI once they do

feel it's safe and secure

everyone so that's why I asked the

question so it's not there yet so it

does it's not representative doesn't

represent everyone in this room um the

world is much more diverse than what we

have in in the room so it doesn't work

so maybe this is where the debate starts

so we're you know open source we want to

be open source want all the you know all

my students are superstars and Geniuses

want all the next generation of the

world to be able to give their

creativity their curiosity because

that's how human flourishing happens but

if we just let the algorithms uh again

on their own I think that we really have

to rethink is it is it where's the

trading data come from where's the

transparency where is the transparency

does it work for all of us I think if uh

those an those questions are answered

well we have a we' have the majority of

folks you know opting in and then and

hopefully making it better right open

sourcing is because you can get all the

good ideas and enhance things so we see

that you know coming enhancing it making

it work for everyone but I think we you

know here and in very intentional

where's the transparency where's the the

trust uh you know has it kind of gotten

away from us so these are really

important questions and Yan I want to

push you one more time and then I really

I hope you all have your questions ready

because I'm coming to you next um I want

to push you one more area which is

values and I wrote about this last year

that you know social media has been

about content moderation what speech do

you allow where do you draw the lines

obviously you know it's something that

meta has spent a lot of time on has had

different approaches um but it strikes

me that these AI systems are going to

have to have values and I wrote that you

know your PC doesn't really have a set

of values your smartphone you know yes

there's some App Store moderations so

you know at the extreme there's some

limits um but the AI system is going to

answer the hard questions and you know

how do we do that in a world where you

know people in the Middle East have

different values than uh people in the

US people in the US have different

values than people in the US um recently

meta made ACH bunch of changes to how

it's going to approach that allowing a

lot more speech even uh that might be

considered very offensive distasteful

even

dehumanizing where is the role of the

tech companies in putting their thumb on

the scale of the values you know I how

much pressure is there going to be from

governments to control what speech how

AI chat Bots for example answer

questions around gender sexuality human

rights so there is a interesting debate

about this so this is not specialy I

should tell you but um but it's an

interesting topic nevertheless that I'm

interested in um

So Meta has gone through several phases

uh concerning content moderation and uh

how how is how best to do it um and uh

including with questions not not just

about toxic content but also about uh

disinformation which is much more

difficult problem to deal with so uh

until 2017 let's say uh detecting things

like hate speech on U on social networks

was very difficult because the

technology just wasn't up to Snuff and

counting on users to flag uh

objectionable content and then have it

reviewed by humans just doesn't scale

particularly if you need those humans to

speak every language in the world um and

so that just was not technologically

possible you just couldn't do it

uh and then what's happened is that

there's been this you know enormous

progress in natural language

understanding uh since 2017 basically

and and that has made enormous amount of

progress so now detecting H speech in

every language in the world is basically

possible with some good level of of

reliability so the proportion of ha

speech for example that is taken down

automatically by AI system was on the

order of 20 to 25% late 207 uh late 2022

5 years later because of Transformers

are supervision you know all the stuff

that that uh uh is everybody is excited

about today uh it was

96% now that probably went too far

because uh the number of false positives

of of of good content that was taken

down is probably pretty high so there

are countries where people just want to

kill each other and you probably want to

kind of you know calm down so so put the

threshold detection threshold pretty low

countries where there is an election and

and you know things going to r r up so

also you want to lower the threshold

detection so that more things get U get

taken down to sort of Camp people down

um but then most of the time you want

people to be able to debate important uh

societal question including for

questions that are you know very

controversial like like gender and and

and political opinions some somewhat

extremes and so what's um what's

happened recently is uh the company

realized it went a little too far and

and there were like just too many Force

positives um and now the the the

detection trols are going to be changed

a little bit to

authorize uh discussions about topics

that are you know big questions of

society even if if the topic is

offensive to some to some people so

that's that's a big change um but it's

it doesn't mean content moderation is

going to go away it's just there it's

just you change the threshold and again

the answer is different in different

countries so uh in Europe it's illegal

hate speech is illegal um you know

neonazi propaganda is illegal right you

you have to do it for legal reason you

have to moderate that for legal Reason

Not So in in the US in various countries

you have different standards as as you

said um then there is a question of

disinformation and there uh until um

until now meta used uh fact checking

organization to fact check the big uh

post that had a lot of uh gathered a lot

of attention but it turns out this

system doesn't work very well it doesn't

scale you don't have a large coverage of

uh of the content that is being posted

because those organization you know

there's only a few of them and they have

a few people working for them and and so

they they can't just uh debunk every you

know uh dangerous misinformation that

circulates on social networks so the

system that is being implemented now

that will be red out is um is qu forcing

essentially have people themselves um um

you know kind of write uh comments uh on

on uh on posts that are controversial um

and that is likely to have much better

coverage there are some studies that

show that this is a a better a better

way of doing conent moderation

particularly if you have some sort of

karma system where people who make

comments that turn out to be reliable or

liked by other people then so that they

get promoted several uh uh forums have

used this system in for many years um so

the The Hope um with the MAA is that

this will actually work better and it

also has a big Advantage which is that U

meta has never seen itself as having the

legitimacy to decide what is right or

wrong for society um and so in the past

has asked governments to to

regulate as governments around the world

this was during the first Trump

Administration uh tell us what is

acceptable on social networks on on on

for online discussion and the answer was

cricket there was basically no answer I

think there was some discussion with the

government in France but uh the Trump

administration at the time the first one

said where the First Amendment here go

away you're on your own um so the all

those those policies kind of resulted

from this absence of uh regulatory

environment um and now it's quite source

is you know content moderation for the

People by the people well there's much

more uh we could talk about but I don't

want to oh yes if I could get us back to

values I think that's that's the right

question so um if we can that's that

should be the first question uh what are

the values so and you have to be able to

articulate your values like articulate

my values it's it's up to leadership to

articulate values so you know for me is

um Integrity Excellence curiosity

community community encompasses

belonging and collaboration so if you

can articulate your values and then as

designers as Builders as technologists

flow from those values we could get it

right what if we get this right so I

think you really we need to back up so

Med I should articulate and in the you

know the checking what are the values do

we have aligned values then we can

collaborate then we can all collaborate

work together and respect um our

cultural differences and all the you

know the Cornucopia that humanity is and

and that's that's wonderful and that's

the opportuni is to go across uh you

know all the cultures but but I think we

fundamentally still have to have the

discussion about values and do we share

values that's that's the I think

fundamental yeah the core core core

share values that you know need to be

expressed I mean the in that sense the

content policy from it are published

right so it's not it's not a secret um

but then there is the implementation of

it right and and and um M the p as made

mistake deploy the system and then

realize that this is not working the way

we wanted it so can of R it back and

replace it by other systems it's it's

constantly but you could lead you could

you could lead in industry and you know

lean in and be out in the front that

discussion uh by all measure actually ma

is is leading in terms of content

moderation absolutely and daa is that

your sense I mean are you concerned with

the new policies that um you know I mean

obviously it's very difficult to say

what are shared values there are a lot

of debates again even in the US at the

same time um you know we talked about a

human centered world and the new

policies um certainly allow a lot of uh

dehumanizing Speech whether it's uh

comparing women to objects uh trans

people to it g people mentally ill have

they gotten that balance right or are

they going no we don't have the right

policies absolutely no emphatically

no we know what's wrong and

right we know human behavior we know

civility we know what makes you happy

when you're teaching your kids we should

probably look at at our our children our

kids and the Young Generation as as well

especially when we talk about um values

and and what we have and and you know

who who we who we aspire to to be

there's a chance to get it right but um

you know we've run the experiment uh you

know internet one internet two I think

we've running the experiment so this is

the opportunity to to get it right I

want to bring in the audience who who

would like to uh build on the discussion

we've had and please just say your name

and where you're from uh there's a mic

coming around but keep the intro short

and ask a question I'm Mukesh from

Bangalore India uh so yeah your group is

at the Forefront of AI research and so

are many other groups around the the

world do we know where we are going like

can is there a mental model for 5 years

from now because we all speculating and

asking questions about where a is today

challenges and so on do we understand

where we're going enough to we have some

prediction about 5 years or is just too

much wide open so my colleagues and I

ADM certainly understand where we are

going I can't claim to understand what

other people are are doing particularly

the ones that are not publishing their

their research and basically you know

have clammed up in in recent times um

but um the way I see things going so

first of all uh I think the the shelf

life of the current Paradigm uh large

language model is fairly short probably

3 to five years I think within five

years nobody in the right mind would use

them anymore at least not as kind of the

central component of an AI system um one

analogy that some people have made which

have uh recycled is um llms are are good

at manipulating language but not at

thinking okay manipulating language is

done by little piece of the brand right

here called the bar area it's about it's

about this big it only popped up in the

last few hundred thousand years can't be

that

complicated what about this the frontal

cortex that's where we think right we we

don't know how to reproduce this um so

that's what we're working on um you know

having systems s of build Mentor models

of the world

so if the plan that we're working on

succeeds you know with the the the

timetable that that we we uh we hope uh

within 3 to 5 years we'll have system

that are complete different Paradigm

they may have some level of Common Sense

they may be able to learn how the world

works from observing uh the world go by

and maybe interacting with it uh you

know deal with uh real world not just

discret discret World um and open the

door to another application I want to

give you just a very U uh interesting um

calculation uh a typical uh fish model

today large language model is train on

20 trillion tokens or 30 trillion tokens

uh a token is typically three bytes so

that's about uh you know 9 10 to the 13

bytes 10 to the 14 bytes okay let's

round it up uh this basically is uh

almost all of publicly available text on

the internet it would take any of us

sever hundred, years to read through it

okay um now compare this with what a

four-year-old has seen in the four years

of life uh you can put a number on how

much information gets to the visual

cortext or or or through touch if you BL

um and it's about um it's about 2 2

megabytes per second about 1 Megabyte

per optic nerve about 1 B per second per

optic nerve fiber we have one million of

them for each eye uh multiply this by

four years and now four in four years um

a child has been awake a total of 16,000

hours so figure out how how many bites

that is 10 to the 14 same number in four

years um so what it tells you is that

we're never going to get to human level

AI

which some people call AI but that's a

misn um we're never going to get to

human level AI by just training on text

we need systems to be able to learn how

the world works from sensory data um and

and so that means LNS are not it talk

about that not we're not going to get

hum within two years like what some of

people have been saying and you've been

talking about that as well yeah so

that's that's my point you know this is

and it's infancy so I think it's

actually you know um that's the way to

clear it where you know LMS are in the

infancy is you know infant I

four-year-old is not infant but very

very early on uh but when you move to

generative biology um training data when

you move to sensors internet of thing

when you move to you know the almost

infinite you know amount of of data

information we have and um you know just

multi- sensory you're talking about you

know you have the the glasses on you

know your vision but you're looking at

text how much do we get tacti hearing

sensing smelling right have you all had

your coffee this morning you know what's

the first thing what was the first thing

that you know you really related to this

morning probably you know breakfast

sense of coffee smell so would put the

multisensory capabilities again for for

humans and I want to be clear from the

earlier you know com Humanity

flourishing humanity and all living and

all living beings all all living the

appreciation for all of life for all of

life human centered design in terms of

some our Technologies but uh you you get

to choose your orientation you get to

choose who you're uh designing for and

so I think that's really important too

not the egocentric Humanity versus the

rest of you know it's it's that's that's

the question you know how long will we

be here um spaceship Earth Technologies

for space up there that's my that's my

specialty um it doesn't need us you know

so a little humility please being being

humble earth going to be fine without

Humanity we're a bit of a nuisance a

huge nuisance so you know Earth is 4.5

billion years old I have my sister

planet Mars probably find past life

there about 300 3.5 billion or so so

again that that view let's please you

know with with humility and approach

this and then the question is you know

do we want to live in Balance do we we

want to live the best lives we can and

flourish and then then I think you just

approach it you know with different

questions you approach solutions from a

from a different perspective thanks I

think I heard something over here I'm

not sure if it was a phone or a question

but I know I see a

hand um talk a lot about existence and

um there's a coming because we have a

live stream audience and say who you are

moris band light speed um for Dava um

you know you talk about AI you talk

about existence to I'm glad you're

making life or working on making life

human life a multiplanetary species um

where does AI fit in into this broader

need um do you see it as an existential

threat do you see it as an existence

enhancing technology for example

generative bio is it our great filter

thank you for the the the question so um

you know I think we're the threat I

think the people are the threat you know

not my not my algorithms uh and for you

know the question when I'm I do think

about you know searching finding life

elsewhere in the universe it's a huge

help so when you say ai ai is not very

useful anymore it's almost just like

saying technology so then we can get now

we should say the the specifics you know

if we're talking um so when it comes to

to travel space for me humans are here

on Earth We're sending our probes and

our scientific instruments so it has a

lot to do with autonomy and autonomous

system and no the human having

information here but that that that Loop

of of information sensing and and

exploration but these are all autonomous

um robots and systems we are going to

send people and then we bring our own

supercomputers with us so that first

human Mission to Mars will be it'll

surpass our current 50 years of

exploring on Mars so that's the benefit

of of humans or you know human intellect

but so so it's a great question so it's

a mix up is a threat we use it to the

advantage of again capabilities

searching exploring and you can get in

my case searching for the evidence of

Bio signatures or or finding life

elsewhere so when you're focused you

know and you know your mission and again

be very transparent about how you're

using algorithms Ai and we always bring

in uh something that's you know very

much uh Missing in in most of the

development when we get down to more

foundational models specific um you know

personalized you know foundational

capability whether it's for health or

climate or exploration you got to bring

in the physics so there physics is more

if you just if you let things go just

mathematically statistically I mean look

at where we're at fantastic but I'm a

big believer and again I'm biomimic

trying to I'm trying to understand

nature I'm trying to understand living

systems always bringing in like

foundational physics with my math and

you know and you proceed along that that

course so while we continue the

discussion in here I also invite those

online we have a couple questions for

you what excites you about the

technology that we're talking about what

worries you and we have the opport to do

some more word cloud so if you're online

and using slido please share your

thoughts there and then we had a

question

there uh they're gonna bring a

microphone everyone if you can just wait

for a mic it'll help those online

Martina hirayama state Secretary for

Education research and Innovation

Switzerland uh my question uh goes to

you Dava so you talk about uh values

concerning AI so we have a divide

concerning access to AI or not what

influence will it have uh if uh we

consider that we do not share the same

values on Earth in all areas where we

live not even talking

about space what influence will this

have on divide yeah it's um so again I

think it's fundamental to so um you know

I give a list of five or six so my hope

is it um we can agree on uh t or three

of those two or you know just two or

three of those you probably won't be uh

the entire set but but I think we have

to look for agreement and shared values

and and then and then work together and

um if not then that's maybe the scenario

that plays out of of the threat division

destruction I don't want that path I

think we have an alternate path so I

think the hard work is People to People

sure policies regulation what do we

agree on what do we agree on how what

what future scenarios and their

scenarios it's very plur what future

scenarios do we agree on and if we can

agree on some of those if we can share

some of those those values and I I think

we can we could take a poll you know see

if we can get one amongst all this you

know diversity here so that's uh you

know it's a it's not an answer it's just

part of the discussion of what can we um

share and what do we what do we share

together and and make that the building

blocks to to get it right and and Yan

that is kind of the challenge of

building these systems for a globe again

where the world doesn't agree on a lot

um there's hopefully some basic things

we agree on though it seems like we

struggle even on those I know you've

talked about using Federated learning

and and to really make sure the world is

represented in these models but how do

we build for a world where there is so

much disagreement again when AI systems

aren't going to just moderate content

they're going to create an answer

content well I I think the answer to

this is diversity so if uh again if you

have two or three AI system that that

all come from the same location you're

not going to get diversity so the only

way to get diversity is having systems

that

are uh you know train on all the all the

languages and cultures and VAR systems

in the world uh and those are Foundation

models and then they can be fine-tuned

by a large uh diversity of of people who

can build assistance with different

ideas of what you know good value

systems are and and then people can

choose so it's the same idea is a

diverse press right you need a diversity

of opinion in the Press um to to at

least have the the basic ingredient of

democracy so it's you know it's the same

for it's going to be the same for AI

system you you need them to be diverse

so one way to do this I mean it's quite

possible that it's quite likely that it

it's going to be very difficult for a

single entity to train a financial model

on all the data all the cultural data in

the world and that may eventually have

to be done in sort of a Federated

fashion or or distributed fashion where

every regions in the world or every

interest group or whatever has their own

Data Center and their own data set and

they contribute to training a a big

Global model uh that may eventually

constitute the repository of all human

knowledge I saw a hand over here and if

you can wait for the mic thanks yeah

well we're passing the mic and I think

that's that's much more exciting uh to

me more the Federated toing it again

transparency because then it's it's it's

more customized it's more personalized

it's going after you know for the work

that it's it's again going after a

medicine or health or a speciic a

specific you know breast you know it can

be more specific and much more precise

so to me that's very

exciting hi my name is MTA josi and I'm

from London um I was listening to a

panel yesterday and they talked about a

concept that really startled me and I

went back and did a bit of research on

it and it's called alignment faking in

llms uh which is about how you know the

llm models are giving answers which are

which they are faking to align to

whatever is being asked to them or

whatever the general can say is probably

an experiment that has happened in the

last few months but it was really

startling and I just thought I'd get a

few thoughts from you on that okay um I

have a perhaps a slightly controversial

opinion about this which is that uh to

some extent llms are intrinsically

unsafe okay because they're not

controllable you don't really have any

direct way of controlling uh whether

what they say is you know certain

characteristics you know with respects

guard rails the only way you can do this

is by training them to do it but of

course that training can be undone by

you know going outside of the the the

domain where where they've been trained

um so to some extent they're inally

unsafe now that's not particularly

dangerous because they're not

particularly smart either right so they

they're useful um they they are in terms

of intelligence they are more like

intellig assistance in the sense that

you know if they produce a text you know

that a lot of it can be wrong in it and

you have to kind of go you know do a

pass on it and correct some of the

mistakes and like you know know what

you're doing it's a bit like you know

driving assistance for cars we don't

have completely autonomous consumer cars

but we have driving assistance and it

works really well so same thing uh but

we should forget about llms so this idea

that somehow we should extrapolate the

uh capability of llms and and realize oh

they can you know fake the intention

first they don't have any intentions and

and like you know simulate values they

don't have any values um and uh and you

know convince people to do horrible

things they don't have any notion of of

of what this is at all um and as I said

they're not going to be with us five

years from now we're going to have much

better system that are objective driven

where the output that those system will

produce will be by by reasoning and the

reasoning will U guarantee that whatever

output is produce satisfy certain guard

rails and the system will not those

system will not be able to be um it

wouldn't be possible to jailbreak to

jailbreak them by changing the promp

basically because that would be sort of

hardwired in the in the guard rails so

given what Yan just said daa you know

the Big Talk the big buzzword this year

is agents and giving more power to these

llms given what Yan just said about

their limitations and this is one of the

companies making it should we be worried

about giving more autonomy and agencies

to A system that has no values makes

mistakes yeah well and I don't think so

I agree with what y said you know that

LM said they're not smart they don't

have rationality they don't have an

intention I mean they're just they're

just lacking think of them as you know

math math and statistical you know

probabilities like that so all of the

probably what we much more care about

you know in humans is well judgment

that's you know the question is like

well this is seems very alerting because

it's you know fakes fake fakes of any

types are are alert right so um the

question what do we do about this

because you know agents so you know

agenic uh you know it is turning into

gentic so simple there's some simple uh

I don't know if there are solutions

there just simple ideas we can do right

um you know we have copyright things and

things like that what if it just you

know comes up every time we're using a

generative uh you know model why is why

isn't it watermarked why isn't you know

to why don't we know that uh you know

what's you know is this coming from a

human is this you know coming from an

algorithm just just you know just

visually just saying you know just

Watermark that you know it's generative

just some more information about what

you're looking at so the person the user

you know if this is being you know

served up to someone that they can take

it I want to do the flip side of this

argument too debate you know with myself

I mean you know published a paper on um

unlocking creativity you know again with

machine learning it's fantastic some

generative capability you have an idea

we have an idea so we just do some

simple brainstorming

and generate again to me I like actually

images uh you know the text because it

it maps to the human brain we're almost

perfect in terms of image mapping and

and looking at visuals so you say my

sentence what's that image and you know

Yan can have has and we look down and

we're going to have a really nice

discussion it's going to help us

actually be more creative more we can

have more discussion if it's kind of a

prompt for us you know that's where it's

a tool you know it really is then an

assistant it's helping us Converse and

have a discussion or or a debate I think

it should definitely be flagged we know

we have to know where it comes from we

have to know you know what the

ingredients are into the recipe so it's

hard to believe we only have a couple

minutes left and I want to give each of

you a chance to give us one thing we

haven't talked about what aren't we

talking about enough that we should be

talking about and maybe we'll be talking

about next year okay I'm going to go by

the list that we're see here exactly

this is what excites you the most about

technology okay brain computer interface

forget about that uh this is not

happening anytime soon at least not the

invasive type that neuralink is working

on the the non-invasive type so things

like you know electrogram bracelets that

m is working on yes that's happening

this year um and that's exciting

actually uh but but like drilling your

brain no um except for clinical purpose

uh gaming virtual world meta of course

has been sort of very active in this in

this space with metaverse say

exploration you are the expert uh it's

exciting as well um regulation uh that's

a very interesting topic um that

um uh I think people are in government

have have been brainwashed to some

extent into believing in the existential

risk story and has led to regulation

that are frankly

counterproductive because the effect

that they have is essentially make uh

open source the distribution of Open

Source AI engine essentially illegal and

in my opion that's way more dangerous

than um than all the other potential

dangers um consumer robotics as I said

maybe the coming decade will be the

decade of uh of of Robotics because

maybe we'll have ai systems that are

sufficiently smart to understand how the

real world works and in your previous

Cloud there was efficiency and and power

uh consumption uh efficiency there is

enormous motivation and uh uh incentive

for the industry to make AI infs more

efficient so you don't have to worry

about people not being motivated enough

to make AI systems efficient this is the

main cost of running an AI system is

power consumption so enormous amount of

work there uh but the technology is what

it is thanks DAV we have a minute left

yeah speed round I'll take I'll take uh

three of them um um I um politely uh

disree brain computer interfaces um no

we it's it's not it's not off it's

happening now um in terms of we have a

digital central nervous system so we're

are already having brain control over

especially in the the area of

breakthrough and Technologies for

replacement for Prosthetics so half

human half robotic new robotic legs you

know uh get rid of phantom phantom foot

because the brain is literally

controlling the robot so it's uh we're

to the the cyborg phase we're doing that

it's it's implanted people are walking

around um soon will'll hopefully be

paraplegics in the future maybe

quadriple so the brain is controlling uh

you know a digital uh Central

nervousness the brain is quite powerful

so it's the surgery so I'd love to talk

about that but but that's here that's a

that's not even the future that's that's

the now um after you know space we

talked about a little bit but again for

scientific purposes uh you know Finding

life what does that why explore out sour

because it tells us it's not option b

sorry Elon it's not option b it's for

flourishing humanity is to appreciate

all of us together our humanity and what

we can get right here on Earth and

definitely living in balance with Earth

so but it's necessary

because when we design for space in the

extreme environments of the Moon Mars

you name it Europa Clipper you anywhere

in the solar system exoplanets it's

because for us it pushes us it pushes

the technology makes me um you know

really sharp in the game and ser as

technology so very optimistic about that

I think we will find the evidence of

life or past life in the next decade

robotics this is um consumer robotics

okay um but what if we what if it's just

the robotic again it Hardware software

robotics should think of you know

physical systems well guess what now

what robots to they are the AI they're

the algorithm they're the software so we

do get to that physical cyber we get to

where we don't talk about hardware and

software we get to know just the

robotics or the machine it's it's

embedded with uh the software I my uses

my favorite um use cases and for health

you know revolutionizing um

individualized you know personalized

medicine things like that rather than

buying it and more stuff and more stuff

and more consuming what if you you make

your own again we're back to open source

let everyone you know do it yourself

make it yourself open source it and use

it from all recycled you know let's

think about you know what's circular so

what can we do with everything any waste

that's the new to me that's uh the new

robotic you know informed physical cyber

system of the future in the hands of

course of our our kids and they'll do

some just a little bit of Education

they'll do some pretty wonderful things

with it if you leave it to the next

Generation Well that's a great place uh

to leave things we are going to have to

leave it there thank you so much David

Newman from MIT Yan Lon from meta

everyone in the room and everyone who's

joined

us thank you