Why AI Is Incredibly Smart and Shockingly Stupid | Yejin Choi | TED

Prevodilac: Milenka Okuka Lektor: Una Kavazović

So I'm excited to share a few spicy thoughts on artificial intelligence.

Uzbuđena sam da podelim nekoliko pikantnih misli o veštačkoj inteligenciji.

But first, let's get philosophical

Prvo pak, bacimo se na filozofiju

by starting with this quote by Voltaire,

počinjući sa ovim citatom Voltera,

an 18th century Enlightenment philosopher,

filozofa prosvetiteljstva iz XVIII veka,

who said, "Common sense is not so common."

koji je rekao: „Zdrav razum nije tako uobičajen.”

Turns out this quote couldn't be more relevant

Ispostavlja se da ovaj citat ne bi mogao biti relevantniji

to artificial intelligence today.

za današnju veštačku inteligenciju.

Despite that, AI is an undeniably powerful tool,

Uprkos tome, VI je nesumnjivo moćno oruđe,

beating the world-class "Go" champion,

pobeđuje šampiona svetske klase u Gou,

acing college admission tests and even passing the bar exam.

ubedljiva je na prijemnim ispitima, pa čak i polaže pravosudni ispit.

I’m a computer scientist of 20 years,

Već 20 godina sam kompjuterska naučnica

and I work on artificial intelligence.

i radim na veštačkoj inteligenciji.

I am here to demystify AI.

Ovde sam da demistifikujem VI.

So AI today is like a Goliath.

Dakle, današnja VI je poput Golijata.

It is literally very, very large.

Bukvalno je veoma, veoma velika.

It is speculated that the recent ones are trained on tens of thousands of GPUs

Spekuliše se da su skorašnje obučene na desetinama hiljada grafičkih procesora

and a trillion words.

i bilionima reči.

Such extreme-scale AI models,

Slični modeli VI ekstremnih razmera,

often referred to as "large language models,"

koji se često označavaju kao „veliki jezički modeli”,

appear to demonstrate sparks of AGI,

čini se da pokazuju iskre VOI,

artificial general intelligence.

veštačke opšte inteligencije.

Except when it makes small, silly mistakes,

Osim kada prave malene, sulude greške,

which it often does.

a to često rade.

Many believe that whatever mistakes AI makes today

Mnogi veruju da kakve god greške da VI danas napravi,

can be easily fixed with brute force,

lako mogu da se poprave sirovom silom,

bigger scale and more resources.

većim razmerama i sa više resursa.

What possibly could go wrong?

Šta bi uopšte moglo da pođe po zlu?

So there are three immediate challenges we face already at the societal level.

Imamo tri trenutna izazova s kojima se već susrećemo na nivou društva.

First, extreme-scale AI models are so expensive to train,

Prvo, ekstremne razmere modela VI su toliko skupe za obučavanje,

and only a few tech companies can afford to do so.

i svega nekoliko tehnoloških kompanija to može da priušti.

So we already see the concentration of power.

Dakle, već svedočimo koncentraciji moći.

But what's worse for AI safety,

Međutim, što je još gore po VI bezbednost,

we are now at the mercy of those few tech companies

trenutno smo prepušteni na milost nekolicine ovih tehnoloških kompanija

because researchers in the larger community

jer istraživači u široj zajednici

do not have the means to truly inspect and dissect these models.

nemaju sredstva da istinski ispitaju i seciraju ove modele.

And let's not forget their massive carbon footprint

I ne zaboravimo njihov ogroman karbonski otisak

and the environmental impact.

i uticaj na okolinu.

And then there are these additional intellectual questions.

A potom imamo dodatna pitanja o inteligenciji.

Can AI, without robust common sense, be truly safe for humanity?

Može li VI, bez razvijenog zdravog razuma, uistinu biti bezbedna za čovečanstvo?

And is brute-force scale really the only way

I da li je upotreba sirove sile u širenju njenih razmera zaista jedini način

and even the correct way to teach AI?

ili uopšte ispravan način za podučavanje VI?

So I’m often asked these days

Ovih dana me često pitaju

whether it's even feasible to do any meaningful research

da li je uopšte izvodljivo obaviti ikakvo smisleno istraživanje

without extreme-scale compute.

bez ekstremnih računarskih performansi.

And I work at a university and nonprofit research institute,

Radim na univerzitetu i neprofitnom istraživačkom institutu,

so I cannot afford a massive GPU farm to create enormous language models.

te ne mogu da priuštim masivne farme GPU kako bih stvorila ogromne jezičke modele.

Nevertheless, I believe that there's so much we need to do

Ipak, verujem da toliko toga moramo da uradimo

and can do to make AI sustainable and humanistic.

i možemo da uradimo kako bi VI bila održiva i humanistička.

We need to make AI smaller, to democratize it.

Moramo da VI učinimo manjom, da je demokratizujemo.

And we need to make AI safer by teaching human norms and values.

I moramo da učinimo VI bezbednijom učeći je ljudskim normama i vrednostima.

Perhaps we can draw an analogy from "David and Goliath,"

Možda možemo da napravimo analogiju sa „Davidom i Golijatom”,

here, Goliath being the extreme-scale language models,

Golijat su ovde jezički modeli ekstremnih razmera,

and seek inspiration from an old-time classic, "The Art of War,"

kao i da potražimo inspiraciju u drevnom klasiku „Umetnost ratovanja”,

which tells us, in my interpretation,

koji nam kaže, prema mom tumačenju,

know your enemy, choose your battles, and innovate your weapons.

upoznajte neprijatelja, birajte bitke i unapređujte oružja.

Let's start with the first, know your enemy,

Započnimo sa prvim, upoznajte neprijatelja,

which means we need to evaluate AI with scrutiny.

to znači moramo pažljivo da procenjujemo VI.

AI is passing the bar exam.

VI polaže pravosudni ispit.

Does that mean that AI is robust at common sense?

Da li to znači da je VI ovladala zdravim razumom?

You might assume so, but you never know.

Mogli biste to da pretpostavite, ali nikad ne znate.

So suppose I left five clothes to dry out in the sun,

Zato pretpostavimo da sam ostavila pet tkanina da se suše na suncu

and it took them five hours to dry completely.

i da im je trebalo pet sati da se skroz osuše.

How long would it take to dry 30 clothes?

Koliko je potrebno da se osuši 30 tkanina?

GPT-4, the newest, greatest AI system says 30 hours.

GPT-4, najnoviji, najbolji sistem VI kaže 30 sati.

Not good.

Nije dobro.

A different one.

Drugi primer.

I have 12-liter jug and six-liter jug,

Imam bokal od 12 i šest litara,

and I want to measure six liters.

i želim da izmerim šest litara.

How do I do it?

Kako da to postignem?

Just use the six liter jug, right?

Prosto uzmete bokal od šest litara, zar ne?

GPT-4 spits out some very elaborate nonsense.

GPT-4 izbacuje istinski krajnje razrađenu besmislicu.

(Laughter)

(Smeh)

Step one, fill the six-liter jug,

Prvi korak, napunite bokal od šest litara,

step two, pour the water from six to 12-liter jug,

drugi korak, prelijte vodu iz bokala od šest u bokal od 12 litara,

step three, fill the six-liter jug again,

treći korak, napunite ponovo bokal od šest litara,

step four, very carefully, pour the water from six to 12-liter jug.

četvrti korak, što opreznije, uspite vodu iz bokala od šest u bokal od 12 litara.

And finally you have six liters of water in the six-liter jug

I konačno imate šest litara vode u bokalu od šest litara

that should be empty by now.

koji bi do sad trebalo da je prazan.

(Laughter)

(Smeh)

OK, one more.

U redu, još jedan.

Would I get a flat tire by bicycling over a bridge

Da li ću probušiti gumu, ako biciklom pređem preko mosta

that is suspended over nails, screws and broken glass?

koji visi iznad eksera, šrafova i polomljenog stakla?

Yes, highly likely, GPT-4 says,

Da, krajnje verovatno, kaže GPT-4,

presumably because it cannot correctly reason

verovatno jer ne može da ispravno rezonuje

that if a bridge is suspended over the broken nails and broken glass,

da ukoliko most visi iznad polomljenih eksera i stakla,

then the surface of the bridge doesn't touch the sharp objects directly.

onda površina mosta direktno ne dotiče oštre objekte.

OK, so how would you feel about an AI lawyer that aced the bar exam

U redu, kako biste se osećali da imate advokata VI

koji je ubedljivo položio pravosudni ispit,

yet randomly fails at such basic common sense?

pa ipak nasumično podbacuje u sličnim osnovama zdravog razuma?

AI today is unbelievably intelligent and then shockingly stupid.

Današnja VI je neverovatno inteligentna, a opet i zapanjujuće glupa.

(Laughter)

(Smeh)

It is an unavoidable side effect of teaching AI through brute-force scale.

Radi se o neizbežnoj nuspojavi obučavanja VI putem sirove sile razmera.

Some scale optimists might say, “Don’t worry about this.

Neke pristalice razmera bi rekli: „Ne brinite zbog ovoga.

All of these can be easily fixed by adding similar examples

Sve ovo lako može da se popravi dodavanjem sličnih primera

as yet more training data for AI."

kao dodatnih podataka za obučavanje VI.”

But the real question is this.

Međutim, istinsko pitanje glasi:

Why should we even do that?

zašto bismo to uopšte radili?

You are able to get the correct answers right away

U stanju ste da momentalno tačno odgovorite

without having to train yourself with similar examples.

bez potrebe da se podučavate na sličnim primerima.

Children do not even read a trillion words

Deca ni ne čitaju bilione reči

to acquire such a basic level of common sense.

kako bi stekli sličan osnovni nivo zdravog razuma.

So this observation leads us to the next wisdom,

Stoga nas ovo zapažanje vodi do sledeće mudrosti,

choose your battles.

birajte bitke.

So what fundamental questions should we ask right now

Dakle, koja bi temeljna pitanja trebalo da trenutno postavljamo

and tackle today

i da se njima bavimo danas

in order to overcome this status quo with extreme-scale AI?

kako bismo prevazišli status kvo kod VI ekstremnih razmera?

I'll say common sense is among the top priorities.

Tvrdila bih da je zdrav razum među glavnim prioritetima.

So common sense has been a long-standing challenge in AI.

Zdrav razum je istrajni izazov za VI.

To explain why, let me draw an analogy to dark matter.

Da bih objasnila zašto, dozvolite da napravim analogiju sa tamnom materijom.

So only five percent of the universe is normal matter

Dakle, svega pet procenata univerzuma je normalna materija

that you can see and interact with,

koju vidite i s kojom interagujete,

and the remaining 95 percent is dark matter and dark energy.

a preostalih 95 procenata je tamna materija i tamna energija.

Dark matter is completely invisible,

Tamna materija je skroz nevidljiva,

but scientists speculate that it's there because it influences the visible world,

ali naučnici spekulišu da postoji jer utiče na vidljivi svet,

even including the trajectory of light.

uključujući čak i putanju svetlosti.

So for language, the normal matter is the visible text,

U slučaju jezika, normalna materija je vidljivi tekst,

and the dark matter is the unspoken rules about how the world works,

a tamna materija su neizreciva pravila o tome kako svet funkcioniše,

including naive physics and folk psychology,

uključujući naivnu fiziku i narodnu psihologiju,

which influence the way people use and interpret language.

koja utiču na to kako ljudi koriste i tumače jezik.

So why is this common sense even important?

Zašto je zdrav razum uopšte važan?

Well, in a famous thought experiment proposed by Nick Bostrom,

Dakle, u čuvenom misaonom eksperimentu, koji je predložio Nik Bostrom,

AI was asked to produce and maximize the paper clips.

od VI je zatraženo da proizvede najveću moguću količinu spajalica.

And that AI decided to kill humans to utilize them as additional resources,

A VI je odlučila da ubije ljude kako bi ih iskoristila kao dodatni resurs,

to turn you into paper clips.

da vas pretvori u spajalice.

Because AI didn't have the basic human understanding about human values.

Jer VI nije imala osnovno ljudsko razumevanje o ljudskim vrednostima.

Now, writing a better objective and equation

Sad, ispisivanje boljeg cilja i jednačine

that explicitly states: “Do not kill humans”

koja eksplicitno navodi: „Ne ubijaj ljude”,

will not work either

takođe neće funkcionisati

because AI might go ahead and kill all the trees,

jer se VI možda odvaži i ubije sve drveće,

thinking that's a perfectly OK thing to do.

misleći da je sasvim u redu da to uradi.

And in fact, there are endless other things

I zapravo, postoji bezbroj drugih stvari

that AI obviously shouldn’t do while maximizing paper clips,

koje VI očito ne bi trebalo da uradi dok maksimalno uvećava broj spajalica,

including: “Don’t spread the fake news,” “Don’t steal,” “Don’t lie,”

uključujući: „Ne širi lažne vesti”, „Ne kradi”, „Ne laži”,

which are all part of our common sense understanding about how the world works.

koje su sve deo našeg zdravorazumskog razumevanja toga kako svet funkcioniše.

However, the AI field for decades has considered common sense

Međutim, zdrav razum se decenijama u oblasti VI smatra

as a nearly impossible challenge.

gotovo nemogućim izazovom.

So much so that when my students and colleagues and I

Toliko da kada smo moji studenti, kolege i ja

started working on it several years ago, we were very much discouraged.

počeli time da se bavimo pre nekoliko godina, prilično smo obeshrabrivani.

We’ve been told that it’s a research topic of ’70s and ’80s;

Rečeno nam je da je to tema za istraživanje iz ’70-ih i ’80-ih;

shouldn’t work on it because it will never work;

ne bi trebalo time da se bavimo jer nikad neće funkcionisati;

in fact, don't even say the word to be taken seriously.

zapravo, ni samu reč ne izgovarajte, ako želite da vas ozbiljno shvate.

Now fast forward to this year,

Sad premotajte do ove godine,

I’m hearing: “Don’t work on it because ChatGPT has almost solved it.”

slušam: „Ne bavite se time jer je to ChatGPT skoro rešio.”

And: “Just scale things up and magic will arise,

I: „Samo povećajte razmere i magija će se desiti,

and nothing else matters.”

a ostalo nije važno.”

So my position is that giving true common sense

Moj stav je da pružanje istinskog zdravog razuma,

human-like robots common sense to AI, is still moonshot.

zdravog razuma čovekolikim robotima i VI, i dalje nalikuje pohodu na mesec.

And you don’t reach to the Moon

A ne dosežete mesec

by making the tallest building in the world one inch taller at a time.

tako što uvećavate najvišu zgradu centimetar po centimetar.

Extreme-scale AI models

Modeli VI ekstremnih razmera

do acquire an ever-more increasing amount of commonsense knowledge,

usvajaju sve veće i veće količine zdravorazumskog znanja,

I'll give you that.

nesumnjivo je tako.

But remember, they still stumble on such trivial problems

Međutim, upamtite, i dalje se spotiču o tako trivijalne probleme

that even children can do.

koje čak i deca znaju da rešavaju.

So AI today is awfully inefficient.

Stoga je današnja VI užasno neefikasna.

And what if there is an alternative path or path yet to be found?

A šta ako postoji alternativni put ili put koji tek treba otkriti?

A path that can build on the advancements of the deep neural networks,

Put koji bi nadograđivao postignuća dubokih neuronskih mreža,

but without going so extreme with the scale.

a da ne mora da bude toliko ekstreman u razmerama.

So this leads us to our final wisdom:

Ovo nas vodi do naše poslednje mudrosti:

innovate your weapons.

unapređujte oružja.

In the modern-day AI context,

U savremenom kontekstu VI

that means innovate your data and algorithms.

to znači unapređujte vaše podatke i algoritme.

OK, so there are, roughly speaking, three types of data

U redu, postoji, ugrubo govoreći, tri tipa podataka

that modern AI is trained on:

na kojima se obučava savremena VI:

raw web data,

sirovi podaci sa interneta,

crafted examples custom developed for AI training,

osmišljeni primeri izrađeni po meri zarad obučavanja VI

and then human judgments,

i potom ljudske procene,

also known as human feedback on AI performance.

takođe poznate kao ljudske povratne informacije na performanse VI.

If the AI is only trained on the first type, raw web data,

Ukoliko se VI jedino obučava na prvom tipu sirovih podataka sa veba,

which is freely available,

koji su lako dostupni,

it's not good because this data is loaded with racism and sexism

nije dobro jer su ti podaci krcati rasizmom, seksizmom

and misinformation.

i lažnim informacijama.

So no matter how much of it you use, garbage in and garbage out.

Te nije važno u kojoj meri ih koristite, smeće ulazi, smeće izlazi.

So the newest, greatest AI systems

Stoga se najnoviji, najbolji sistemi VI

are now powered with the second and third types of data

trenutno opskrbljavaju drugim i trećim tipom podataka

that are crafted and judged by human workers.

koje osmišljavaju i prosuđuju ljudski radnici.

It's analogous to writing specialized textbooks for AI to study from

Analogno je pisanju specijalizovanih udžbenika za VI da iz njih uči,

and then hiring human tutors to give constant feedback to AI.

a potom unajmljivanju ljudskih instruktora da stalno daju povratne informacije VI.

These are proprietary data, by and large,

Radi se sveukupno o autorskim podacima,

speculated to cost tens of millions of dollars.

za koje se spekuliše da koštaju desetine miliona dolara.

We don't know what's in this,

Ne znamo šta je u njima,

but it should be open and publicly available

ali trebalo bi da budu otvoreni i javno dostupni

so that we can inspect and ensure [it supports] diverse norms and values.

da bismo mogli da ih ispitamo i obezbedimo različitost u normama i vrednostima.

So for this reason, my teams at UW and AI2

Iz tog razloga, moje ekipe sa Univerziteta u Vašingtonu i AI2

have been working on commonsense knowledge graphs

rade na grafikonima zdravorazumskog znanja

as well as moral norm repositories

kao i na repozitorijumima moralnih normi

to teach AI basic commonsense norms and morals.

kako bismo podučavali VI zdravorazumskim normama i moralu.

Naši podaci su u potpunosti otvoreni kako bi bilo ko mogao da ispita sadržaj

Our data is fully open so that anybody can inspect the content

and make corrections as needed

i po potrebi napravi ispravke

because transparency is the key for such an important research topic.

jer je transparentnost ključna za tako važnu istraživačku temu.

Now let's think about learning algorithms.

Sada, razmislimo o algoritmima za učenje.

No matter how amazing large language models are,

Bez obzira na to koliko su sjajni veliki jezički modeli,

by design

po svom sklopu

they may not be the best suited to serve as reliable knowledge models.

možda nisu baš najpodesniji da služe kao pouzdani modeli znanja.

And these language models do acquire a vast amount of knowledge,

A ovi jezički modeli usvajaju ogromne količine znanja,

but they do so as a byproduct as opposed to direct learning objective.

ali postižu to kao nusproizvod nasuprot direktnom cilju učenja.

Resulting in unwanted side effects such as hallucinated effects

A to rezultira neželjenim posledicama poput halucinirajućih efekata

and lack of common sense.

i odsustva zdravog razuma.

Now, in contrast,

Sad, nasuprot tome,

human learning is never about predicting which word comes next,

kod ljudskog učenja se nikad ne radi o predviđanju koja reč sledi,

but it's really about making sense of the world

već se uistinu radi o razumevanju sveta

and learning how the world works.

i učenju kako svet funkcioniše.

Maybe AI should be taught that way as well.

Možda bi i VI trebalo tako podučavati.

So as a quest toward more direct commonsense knowledge acquisition,

Stoga, u pohodu na usvajanje direktnijeg zdravorazumskog znanja

my team has been investigating potential new algorithms,

moja ekipa je istraživala potencijalno nove algoritme,

including symbolic knowledge distillation

uključujući destilaciju simboličkog znanja

that can take a very large language model as shown here

koji mogu da uzmu izuzetno velike jezičke modele, kao što su ovde prikazani,

that I couldn't fit into the screen because it's too large,

koji nisu mogli da stanu na ekran jer su suviše veliki,

and crunch that down to much smaller commonsense models

i sažmu to u daleko manje zdravorazumske modele

using deep neural networks.

upotrebom dubokih neuronskih mreža.

And in doing so, we also generate, algorithmically, human-inspectable,

A radeći to, takođe stvaramo algoritamski i ljudski ispitljive

symbolic, commonsense knowledge representation,

simboličke reprezente zdravorazumskog znanja

da bi ih ljudi mogli pregledati i napraviti ispravke,

so that people can inspect and make corrections

and even use it to train other neural commonsense models.

pa čak i koristiti da obučavaju druge zdravorazumske neuronske modele.

More broadly,

Uopštenije,

we have been tackling this seemingly impossible giant puzzle

bavili smo se ovom, na prvi pogled, nemogućom džinovskom slagalicom

of common sense, ranging from physical,

zdravog razuma, koja seže od psihološkog,

social and visual common sense

društvenog i vizuelnog zdravog razuma

to theory of minds, norms and morals.

do teorije umova, normi i morala.

Each individual piece may seem quirky and incomplete,

Svaki pojedinačni deo može izgledati nezgrapno i nedovršeno,

but when you step back,

ali kada se udaljite,

it's almost as if these pieces weave together into a tapestry

gotovo da ovi delovi zajedno pletu tapiseriju

that we call human experience and common sense.

koju nazivamo ljudskim iskustvom i zdravim razumom.

We're now entering a new era

Ulazimo u novo doba

in which AI is almost like a new intellectual species

u kome je VI gotovo poput nove intelektualne vrste

with unique strengths and weaknesses compared to humans.

sa jedinstvenim snagama i slabostima u poređenju s ljudima.

In order to make this powerful AI

Kako bismo ovu moćnu VI učinili

sustainable and humanistic,

održivom i humanističkom,

we need to teach AI common sense, norms and values.

moramo da podučavamo VI zdravom razumu, normama i vrednostima.

Thank you.

Hvala.

(Applause)

(Aplauz)

Chris Anderson: Look at that.

Kris Anderson: Pazi to.

Yejin, please stay one sec.

Jeđin, molim te ostani za sekund.

This is so interesting,

Ovo je tako interesantno,

this idea of common sense.

ova ideja zdravog razuma.

We obviously all really want this from whatever's coming.

Svi to očito želimo odakle god da dolazi.

But help me understand.

Pomozi mi pak da shvatim.

Like, so we've had this model of a child learning.

Dakle, imali smo taj model dečjeg učenja.

How does a child gain common sense

Kako dete usvaja zdrav razum

apart from the accumulation of more input

osim akumulacije sve više unosa

and some, you know, human feedback?

i nekakvih, znaš, ljudskih povratnih informacija?

What else is there?

Šta još imamo tu?

Yejin Choi: So fundamentally, there are several things missing,

Jeđin Čoj: Dakle, temeljno nedostaje nekoliko stvari,

but one of them is, for example,

ali jedna od njih je, na primer,

the ability to make hypothesis and make experiments,

sposobnost pravljenja hipoteze i vršenja eksperimenata,

interact with the world and develop this hypothesis.

interagovanja sa svetom i razvijanja ove hipoteze.

We abstract away the concepts about how the world works,

Apstrahujemo koncepte o tome kako svet funkcioniše,

and then that's how we truly learn,

a onda tako istinski učimo,

as opposed to today's language model.

nasuprot trenutnim jezičkim modelima.

Some of them is really not there quite yet.

Neki od njih uistinu nisu još ni blizu cilja.

CA: You use the analogy that we can’t get to the Moon

KA: Koristiš analogiju da ne možemo stići na Mesec

by extending a building a foot at a time.

uvećavajući zgradu pedalj po pedalj.

But the experience that most of us have had

Međutim, iskustvo koje većina nas ima

of these language models is not a foot at a time.

o ovim jezičkim modelima nije izraženo u pedljima.

It's like, the sort of, breathtaking acceleration.

Radi se o nekakvom ubrzanju koje oduzima dah.

Are you sure that given the pace at which those things are going,

Jesi li sigurna da imajući u vidu ritam kojim se ove stvari kreću,

each next level seems to be bringing with it

svaki novi nivo izgleda kao da sa sobom nosi

what feels kind of like wisdom and knowledge.

nešto što naliči mudrosti i znanju.

YC: I totally agree that it's remarkable how much this scaling things up

JČ: U potpunosti se slažem da je izvanredno koliko uvećavanje razmera ovoga

really enhances the performance across the board.

uistinu uvećava sveukupne performanse.

So there's real learning happening

Dakle, dešava se učenje

due to the scale of the compute and data.

zahvaljujući razmerama računarskih performansi i podacima.

However, there's a quality of learning that is still not quite there.

Međutim, određeni kvalitet učenja i dalje nedostaje.

And the thing is,

A radi se o tome

we don't yet know whether we can fully get there or not

da mi još uvek ne znamo da li to možemo u potpunosti postići ili ne

just by scaling things up.

pukim uvećavanjem razmera.

And if we cannot, then there's this question of what else?

A ako ne možemo, sledi pitanje šta onda?

And then even if we could,

A potom ako možemo,

do we like this idea of having very, very extreme-scale AI models

da li nam se sviđa ideja da imamo VI modele veoma, veoma ekstremnih razmera

that only a few can create and own?

koje može da stvara i poseduje svega nekolicina?

CA: I mean, if OpenAI said, you know, "We're interested in your work,

KA: Mislim, kada bi iz OpenAI rekli: „Zainteresovani smo za tvoj rad,

we would like you to help improve our model,"

voleli bismo da nam pomogneš da unapredimo naše modele”,

can you see any way of combining what you're doing

vidiš li neki način kombinovanja onoga što radiš

with what they have built?

sa onim što su oni sagradili?

YC: Certainly what I envision

JČ: Zasigurno, ono što zamišljam

will need to build on the advancements of deep neural networks.

će morati da se nadogradi na unapređenja dubokih neuronskih mreža.

And it might be that there’s some scale Goldilocks Zone,

I možda se ispostavi da postoji zona Zlatokose u razmerama

such that ...

takva da...

I'm not imagining that the smaller is the better either, by the way.

Ni ne zamišljam da je manje bolje, usput.

It's likely that there's right amount of scale, but beyond that,

Verovatno da postoji tačna veličina razmera, ali preko toga,

the winning recipe might be something else.

pobednička formula bi mogla da bude nešto drugo.

So some synthesis of ideas will be critical here.

Dakle, nekakva sinteza ideja će da bude ključna ovde.

CA: Yejin Choi, thank you so much for your talk.

KA: Jeđin Čoj, mnogo ti hvala na tvom govoru.

(Applause)

(Aplauz)