The Inside Story of ChatGPT’s Astonishing Potential | Greg Brockman | TED

Prevodilac: Strahinja Tomic Lektor: Milenka Okuka

We started OpenAI seven years ago

Osnovali smo OpenAI pre sedam godina

because we felt like something really interesting was happening in AI

jer smo osetili da se nešto veoma zanimljivio dešava sa VI

and we wanted to help steer it in a positive direction.

i želeli smo da to usmerimo u pozitivnom smeru.

It's honestly just really amazing to see

Iskreno, neverovatno je gledati

how far this whole field has come since then.

dokle se stiglo u ovoj oblasti.

And it's really gratifying to hear from people like Raymond

Zadovoljavajuće je kada nam se jave ljudi poput Rejmonda

who are using the technology we are building, and others,

koji koriste tehnologiju koju pravimo, kao i drugi, za mnoštvo divnih stvari.

for so many wonderful things.

We hear from people who are excited,

Jave nam se ljudi koji su uzbuđeni,

we hear from people who are concerned,

jave nam se ljudi koji su zabrinuti,

we hear from people who feel both those emotions at once.

jave nam se i ljudi koji su i jedno i drugo istovremeno.

And honestly, that's how we feel.

I iekreno, tako se i mi osećamo.

Above all, it feels like we're entering an historic period right now

Nadasve, izgleda kao da sada ulazimo u istorijski period

where we as a world are going to define a technology

gde mi kao svet definišemo tehnologiju

that will be so important for our society going forward.

koja će biti od važnosti za društvo u budućnosti.

And I believe that we can manage this for good.

I verujem da ovim možemo da upravljamo za dobrobit svih.

So today, I want to show you the current state of that technology

Danas želim da vam pokažem trenutno stanje te tehnologije,

and some of the underlying design principles that we hold dear.

kao i neke osnovne principe dizajna kojih se držimo.

So the first thing I'm going to show you

Prva stvar koju ću vam pokazati je kako izgleda pravljenje alata za VI

is what it's like to build a tool for an AI

rather than building it for a human.

naspram pravljenja tog alata za ljude.

So we have a new DALL-E model, which generates images,

Dakle, imamo novi DALL-E model, koji proizvodi slike,

and we are exposing it as an app for ChatGPT to use on your behalf.

i mi ga kao aplikaciju izlažemo ChatGTP-u da ga koristi u vaše ime.

And you can do things like ask, you know,

I možete raditi stvari, npr. da pitate:

suggest a nice post-TED meal and draw a picture of it.

“Predloži mi fin obrok za posle TED govora i izradi sliku toga.”

(Laughter)

(Smeh)

Now you get all of the, sort of, ideation and creative back-and-forth

Dobićete pregled idejnog i kreativnog procesa, na neki način,

and taking care of the details for you that you get out of ChatGPT.

i detalji koje biste dobili iz ChatGP-a će vam biti sređeni.

And here we go, it's not just the idea for the meal,

I krećemo. Vidite da to nije samo ideja o obroku, nego veoma detaljan pregled.

but a very, very detailed spread.

So let's see what we're going to get.

Pa pogledajmo šta ćemo dobiti.

But ChatGPT doesn't just generate images in this case --

Ali ChatGPT ne proizvodi samo slike u ovom slučaju -

sorry, it doesn't generate text, it also generates an image.

izvinjavam se, ne proizvodi samo tekst, nego proizvodi i sliku.

And that is something that really expands the power

A to je nešto što stvarno širi mogućnosti onoga što za vas može da učini,

of what it can do on your behalf in terms of carrying out your intent.

u smislu sprovođenja vaše namere.

And I'll point out, this is all a live demo.

Napominjem, sve je ovo demonstracija uživo.

This is all generated by the AI as we speak.

Sve ovo proizvodi VI dok razgovaramo.

So I actually don't even know what we're going to see.

Tako da ja zapravo ne znam šta ćemo da vidimo.

This looks wonderful.

Ovo izgleda divno.

(Applause)

(Aplauz)

I'm getting hungry just looking at it.

Ogladniću samo gledajući u to.

Now we've extended ChatGPT with other tools too,

Opremili smo ChatGPT i drugim alatima, na primer, memorijom.

for example, memory.

You can say "save this for later."

Možete reći: “Sačuvaj mi ovo za kasnije.”

And the interesting thing about these tools

A zanimljiva stvar sa ovim alatima je da su svi veoma pregledni.

is they're very inspectable.

So you get this little pop up here that says "use the DALL-E app."

Ovde će iskočiti obavest da “koristite DALL-E aplikaciju”.

A usput, ovo vam dolazi, korisnicima ChatGPT, u sledećim mesecima.

And by the way, this is coming to you, all ChatGPT users, over upcoming months.

I možete pogledati ispod haube i videti šta je zapravo učinio,

And you can look under the hood and see that what it actually did

a to je da je napisao upit kao što bi to učinio i čovek.

was write a prompt just like a human could.

And so you sort of have this ability to inspect

Stoga, na neki način, imate mogućnost uvida u to

how the machine is using these tools,

kako mašina koristi ove alate, što nam omogućava davanje povratne informacije.

which allows us to provide feedback to them.

Now it's saved for later,

Sačuvano je za kasnije,

and let me show you what it's like to use that information

a sada ću vam pokazati kako izgleda upotreba te informacije,

and to integrate with other applications too.

kao i integracija sa drugim aplikacijama.

You can say,

Možete reći:

“Now make a shopping list for the tasty thing

“Napravi sad spisak za kupovinu za onu ukusnu stvar

I was suggesting earlier.”

koja mi je ranije predložena.”

And make it a little tricky for the AI.

I da to sve otežamo za VI.

"And tweet it out for all the TED viewers out there."

“I tvituj to svim gledaocima TED govora.”

(Laughter)

(Smeh)

So if you do make this wonderful, wonderful meal,

Ako budete pravili ovaj prekrasan obrok, definitivno želim da znam kakvog je ukusa.

I definitely want to know how it tastes.

But you can see that ChatGPT is selecting all these different tools

Međutim, možete da vidite da ChatGPT bira sve ove različite alate

without me having to tell it explicitly which ones to use in any situation.

bez izričite naredbe koji da koristi u kojoj situaciji.

And this, I think, shows a new way of thinking about the user interface.

Radeći to, pokazuje nov način razmišljanja o korisničkom intefejsu.

Like, we are so used to thinking of, well, we have these apps,

Navikli smo na način razmišljanja,

imamo aplikacije, prebacujemo se, kopiramo i lepimo iz jedne u drugu,

we click between them, we copy/paste between them,

Obično je to ugodno iskustvo u aplikaciji dok god otprilike znamo menije i opcije.

and usually it's a great experience within an app

as long as you kind of know the menus and know all the options.

Da, volio bih to.

Yes, I would like you to.

Yes, please.

Da, molim te. Uvek je lepo biti učtiv.

Always good to be polite.

(Laughter)

(Smeh)

And by having this unified language interface on top of tools,

I time što su svi ti alati objedinjeni jezičkim intefejsom,

the AI is able to sort of take away all those details from you.

VI može da, na neki način, zaključi od vas sve te detalje.

So you don't have to be the one

Tako da ne morate da crtate svaki sitni detalj onoga što bi trebalo da se desi.

who spells out every single sort of little piece

of what's supposed to happen.

And as I said, this is a live demo,

Kao što rekoh, ovo je demonstracija uživo, tako da se nešto neočekivano može desiti.

so sometimes the unexpected will happen to us.

But let's take a look at the Instacart shopping list while we're at it.

Ali pogledajmo sad onaj Instakart spisak za kupovinu, kad smo već kod toga.

And you can see we sent a list of ingredients to Instacart.

Možete videti da smo poslali spisak namirnica na Instakart.

Here's everything you need.

Tu je sve što vam treba.

And the thing that's really interesting

Zanimljiva stvar je da je tradicionalni korisnički intefejs i dalje bitan, zar ne?

is that the traditional UI is still very valuable, right?

If you look at this,

Ako ga pogledate, možete proći kroz spisak i prilagoditi potrebne količine.

you still can click through it and sort of modify the actual quantities.

And that's something that I think shows

Mislim da pokazuje da tradicionalni korisnički interfejsi ne idu nigde.

that they're not going away, traditional UIs.

It's just we have a new, augmented way to build them.

Stvar je da sada imamo novi, bolji način da ih napravimo.

And now we have a tweet that's been drafted for our review,

Sada dobijamo nacrt tvita da pregledamo pre slanja, što je takođe veoma bitno.

which is also a very important thing.

We can click “run,” and there we are, we’re the manager, we’re able to inspect,

Možemo da kliknemo na pokreni, i eto ga. Mi smo šefovi, možemo da ispitamo,

we're able to change the work of the AI if we want to.

možemo da izmenimo rad VI ako to želimo.

And so after this talk, you will be able to access this yourself.

I nakon ovog govora, vi ćete moći ovome i sami pristupiti.

And there we go.

I eto ga.

Cool.

Strava.

Thank you, everyone.

Hvala svima.

(Applause)

(Aplauz)

So we’ll cut back to the slides.

Vratimo se na slajdove.

Now, the important thing about how we build this,

E sad, važna stvar o tome kako gradimo ove alate,

it's not just about building these tools.

je da se ne radi samo o izradi ovih alata.

It's about teaching the AI how to use them.

Radi se i o podučavanju VI kako da ih koristi.

Like, what do we even want it to do

Šta zapravo želimo da VI uradi

when we ask these very high-level questions?

kada joj postavimo ova pitanja veoma visokog nivoa?

And to do this, we use an old idea.

A da bismo to postigli, koristili smo staru ideju.

Ako se vratimo na rad Alana Tjuringa iz 1950, na Tjuringovom testu, on kaže

If you go back to Alan Turing's 1950 paper on the Turing test, he says,

you'll never program an answer to this.

mašinu nikada nećete moći programirati za ovakve odgovore, ali je možete podučiti.

Instead, you can learn it.

You could build a machine, like a human child,

Možete napraviti mašinu nalik na ljudsko dete,

and then teach it through feedback.

i onda ga učiti kroz povratne informacije.

Have a human teacher who provides rewards and punishments

Neka je ljudski učitelj kažnjava i nagrađuje dok isprobava stvari,

as it tries things out and does things that are either good or bad.

dok radi dobre ili loše stvari.

And this is exactly how we train ChatGPT.

I upravo ovako mi treniramo ChatGPT.

It's a two-step process.

To je proces u dva koraka.

First, we produce what Turing would have called a child machine

Prvo, pravimo ono što je Tjuring nazivao mašinom detetom.

through an unsupervised learning process.

Kroz postupak učenja bez nadozora pokažemo mu čitav svet, čitav internet, i kažemo:

We just show it the whole world, the whole internet

and say, “Predict what comes next in text you’ve never seen before.”

“Predvidi šta ide sledeće u do sad nepoznatom tekstu.”

I ovaj postupak je prožima mnogim divnim veštinama.

And this process imbues it with all sorts of wonderful skills.

Na primer, ako joj pokažete matematički problem

For example, if you're shown a math problem,

the only way to actually complete that math problem,

jedini način na koji će ga rešiti,

to say what comes next,

da kaže šta dolazi posle, ova zelena devetka gore, je da reši taj problem.

that green nine up there,

is to actually solve the math problem.

But we actually have to do a second step, too,

Mi ipak moramo da uradimo i drugi korak,

which is to teach the AI what to do with those skills.

a to je da naučimo VI šta da radi s tim veštinama.

I zbog toga, dajemo povratne informacije.

And for this, we provide feedback.

We have the AI try out multiple things, give us multiple suggestions,

Damo VI da isproba različite stvari, da nam da više predloga,

and then a human rates them, says “This one’s better than that one.”

a onda ih čovek ocenjuje, govoreći: “Ova je bolja od one.”

Ovo ne samo da učvršćuje jednu specifičnu stvar koju je VI rekla,

And this reinforces not just the specific thing that the AI said,

but very importantly, the whole process that the AI used to produce that answer.

nego i čitav proces kojim je VI došla do odgovora.

Ovo joj dozvoljava da generalizuje,

And this allows it to generalize.

It allows it to teach, to sort of infer your intent

da nauči, da na neki način zaključi vašu nameru i primeni je

and apply it in scenarios that it hasn't seen before,

na scenarije koje još nije videla, za koje nema povratnu informaciju.

that it hasn't received feedback.

Now, sometimes the things we have to teach the AI

Sad, nekada moramo da naučimo VI stvari koje ne biste očekivali.

are not what you'd expect.

For example, when we first showed GPT-4 to Khan Academy,

Npr., kada smo prvi put pokazali GPT-4 Akademiji Kan,

they said, "Wow, this is so great,

rekli su: “Opa, ovo je odlično,

We're going to be able to teach students wonderful things.

ovim ćemo moći da naučimo naše studente divne stvari.”

Only one problem, it doesn't double-check students' math.

Problem je bio, model nije proveravao matematiku studenata.

If there's some bad math in there,

Ako je matematika loša, pretvaraće se da je 1 + 1 = 3 i raditi na osnovu toga.

it will happily pretend that one plus one equals three and run with it."

So we had to collect some feedback data.

Tako da smo morali da prikupimo povratne informacije.

Sal Khan himself was very kind

Sal Kan lično je bio veoma ljubazan i ponudio nam je 20 sati svog vremena

and offered 20 hours of his own time to provide feedback to the machine

da pruži povratne informacije mašini radeći uz naš tim.

alongside our team.

And over the course of a couple of months we were able to teach the AI that,

I kroz nekoliko meseci, uspeli smo da naučimo VI sledeće:

"Hey, you really should push back on humans

“Hej, u ovom specifičnom scenariju, trebalo bi da se suprotstaviš ljudima.”

in this specific kind of scenario."

And we've actually made lots and lots of improvements to the models this way.

I na ovaj način smo zapravo uveliko poboljšali model.

And when you push that thumbs down in ChatGPT,

A kada pritisnete onaj “palac dole” u ChatGPT-u,

that actually is kind of like sending up a bat signal to our team to say,

to je kao da pošaljete bat-signal našem timu i kažete:

“Evo slabije oblasti gde vam treba još povratnih informacija.”

“Here’s an area of weakness where you should gather feedback.”

Kada tako radite, to je jedan od načina na koji stvarno slušamo korisnike,

And so when you do that,

that's one way that we really listen to our users

and make sure we're building something that's more useful for everyone.

i brinemo se da gradimo nešto što je korisno svima.

Now, providing high-quality feedback is a hard thing.

Sada, dobijanje kvalitetne povratne informacije je teško.

If you think about asking a kid to clean their room,

Kada pitate dete da spremi svoju sobu,

if all you're doing is inspecting the floor,

ako samo proveravate kakav je pod,

you don't know if you're just teaching them to stuff all the toys in the closet.

ne znate da li ga samo učite da natrpa sve igračke u ormar.

This is a nice DALL-E-generated image, by the way.

Inače, ovo je divna slika koju je napravio DALL-E.

And the same sort of reasoning applies to AI.

Ista logika važi i za VI.

As we move to harder tasks,

Kako idemo ka težim zadacima,

we will have to scale our ability to provide high-quality feedback.

moramo srazmerno poboljšati i mogućnost davanja kvalitetnih povratnih infromacija.

But for this, the AI itself is happy to help.

Ali za ovo, sama VI rado pomaže, kako da joj pružimo bolje povratne informacije

It's happy to help us provide even better feedback

i srazmerno je nadgledamo kako vreme odmiče.

and to scale our ability to supervise the machine as time goes on.

And let me show you what I mean.

Dozvolite da vam pokažem na šta mislim.

For example, you can ask GPT-4 a question like this,

Npr., možete pitati GPT-4 ovako nešto,

of how much time passed between these two foundational blogs

koliko je prošlo vremena od ova dva osnivačka bloga

on unsupervised learning

o učenju bez nadzora i učenju na osnovu ljudske povratne informacije.

and learning from human feedback.

And the model says two months passed.

I model kaže da je prošlo dva meseca.

Međutim, da li je to tačno?

But is it true?

Like, these models are not 100-percent reliable,

Ovi modeli nisu sto posto pouzdani,

although they’re getting better every time we provide some feedback.

iako su sve bolji posle svake povratne informacije koju im damo.

But we can actually use the AI to fact-check.

Međutim, možemo iskoristiti VI da proveri činjenice,

And it can actually check its own work.

i može samu sebe da proveri.

You can say, fact-check this for me.

Možete reći: “Proveri mi ovu činjenicu.”

Now, in this case, I've actually given the AI a new tool.

U ovom slučaju sam VI dao novi alat.

This one is a browsing tool

Ovo je alatka za pretraživanje,

where the model can issue search queries and click into web pages.

gde model može slati upite i pregledati veb-stranice.

Pri tome, ispisuje čitav sled razmišljanja dok to radi.

And it actually writes out its whole chain of thought as it does it.

It says, I’m just going to search for this and it actually does the search.

Kaže da će pretraživati ovo i onda radi pretragu.

Onda nalazi datum objave i rezultate pretrage.

It then it finds the publication date and the search results.

It then is issuing another search query.

A onda radi još jedan upit. Kliknuće na ovu objavu bloga.

It's going to click into the blog post.

And all of this you could do, but it’s a very tedious task.

I kao što vidite, sve ovo možete i sami uraditi, ali je veoma zamorno.

It's not a thing that humans really want to do.

Nije nešto što ljudi stvarno žele raditi.

It's much more fun to be in the driver's seat,

Zabavnije je držati uzde, biti menadžer, gde možete, ako to želite,

to be in this manager's position where you can, if you want,

i tri puta proveriti njen rad.

triple-check the work.

I evo dolazi citat,

And out come citations

so you can actually go

pa možete i da s lakoćom potvrdite bilo koji korak u ovom sledu razmišljanja.

and very easily verify any piece of this whole chain of reasoning.

And it actually turns out two months was wrong.

Ispostaviće se da je dva meseca pogrešno. Dva meseca i jedna sedmica je ispravno.

Two months and one week,

that was correct.

(Applause)

(Aplauz)

And we'll cut back to the side.

Vratićemo se na slajd.

And so thing that's so interesting to me about this whole process

Ono što je meni najinteresantnije kod čitavog ovog procesa

is that it’s this many-step collaboration between a human and an AI.

je da je to saradnja čoveka i VI na više nivoa.

Because a human, using this fact-checking tool

Jer čovek koji koristi ovaj alat za proveru činjenica radi to radi podataka

is doing it in order to produce data

for another AI to become more useful to a human.

koje će da koristi druga VI kako bi postala korisnija za čoveka.

And I think this really shows the shape of something

Mislim da ovo poprima oblik nečega

that we should expect to be much more common in the future,

što možemo očekivati da bude sve učestalije u budućnosti.

where we have humans and machines kind of very carefully

Gde ćemo imati čoveka s jedne strane, a s druge pažljivo i precizno podešenu mašinu

and delicately designed in how they fit into a problem

koja je napravljena da rešava problem na način na koji mi želimo.

and how we want to solve that problem.

We make sure that the humans are providing the management, the oversight,

Mi se staramo o nadzoru, upravljanju, povratnim informacijama,

the feedback,

a mašine rade na pregledan i pouzdan način.

and the machines are operating in a way that's inspectable

and trustworthy.

I zajedno smo u stanju da napravimo još pouzdanije mašine.

And together we're able to actually create even more trustworthy machines.

Mislim da ćemo vremenom, ako usavršimo proces kako treba,

And I think that over time, if we get this process right,

biti u stanju da rešavamo nemoguće probleme.

we will be able to solve impossible problems.

I da vam dam prestavu o koliko nemogućim problemima govorimo,

And to give you a sense of just how impossible I'm talking,

I think we're going to be able to rethink almost every aspect

mislim da ćemo moći da preispitamo gotovo svaki aspekt interakcije sa računarima.

of how we interact with computers.

For example, think about spreadsheets.

Npr., razmislite o tabelarnim prikazima.

They've been around in some form since, we'll say, 40 years ago with VisiCalc.

Prisutni su u nekom obliku već nekih četrdesetak godina, od VisiCalc-a.

I don't think they've really changed that much in that time.

Ne mislim da su se nešto bitnije menjali u tom periodu.

And here is a specific spreadsheet of all the AI papers on the arXiv

A evo posebne tablice o svim radovima o VI na arXiv u poslednjih 30 godina.

for the past 30 years.

There's about 167,000 of them.

Ima ih oko 167 000, kao što možete da vidite iz podataka ovde.

And you can see there the data right here.

But let me show you the ChatGPT take on how to analyze a data set like this.

Ali da vam pokažem kako ChatGTP pristupa analizi ovakvog skupa podataka.

So we can give ChatGPT access to yet another tool,

Možemo mu dati pristup još jednom alatu, Pajton interpreteru,

this one a Python interpreter,

so it’s able to run code, just like a data scientist would.

pa će moći da pokreće kod, kao što bi to radio i naučnik za podatke.

And so you can just literally upload a file

Tako da doslovno možete da učitate fajl i upitate VI o tome.

and ask questions about it.

And very helpfully, you know, it knows the name of the file and it's like,

I uz malo sreće, prepoznaće naziv i tip datoteke i reći će:

"Oh, this is CSV," comma-separated value file,

“Aha, ovo je CSV fajl, tu su vrednosti odvojene zarezom, raščlaniću to za tebe.”

"I'll parse it for you."

The only information here is the name of the file,

Jedine informacije ovde su ime fajla,

the column names like you saw and then the actual data.

imena kolona, kao što ste videli, i na kraju sami podaci.

And from that it's able to infer what these columns actually mean.

I iz svega toga, uspela je da izvede zaključak šta te kolone zapravo znače.

Like, that semantic information wasn't in there.

Semantičke informacije nisu bile tu.

It has to sort of, put together its world knowledge of knowing that,

Na neki način, ona mora na osnovu interdisciplinarnog znanja da poveže:

“Oh yeah, arXiv is a site that people submit papers

“Da, arXiv je sajt gde ljudi dostavljaju svoje radove, pa to mora da su ovi podaci,

and therefore that's what these things are and that these are integer values

a onda ovi celi brojevi predstavljaju broj autora u radu.”

and so therefore it's a number of authors in the paper,"

Sve ovo je posao koji inače rade ljudi, a VI će rado pomoći s tim.

like all of that, that’s work for a human to do,

and the AI is happy to help with it.

Sad, ja ni ne znam šta bih pitao. Srećom, možete pitati mašinu:

Now I don't even know what I want to ask.

So fortunately, you can ask the machine,

“Možeš li napraviti neke istraživačke grafikone?”

"Can you make some exploratory graphs?"

And once again, this is a super high-level instruction with lots of intent behind it.

I opet, ovo je jako napredna i detaljna instrukcija koju prati jako puno namere.

But I don't even know what I want.

Ali ni ja ne znam šta hoću, A VI sad mora da zaključi šta bi mene moglo zanimati.

And the AI kind of has to infer what I might be interested in.

I tako dolazi do nekih dobrih ideja, mislim.

And so it comes up with some good ideas, I think.

Izbacuje histogram o broju autora po radu, vremensku seriju radova po godini,

So a histogram of the number of authors per paper,

time series of papers per year, word cloud of the paper titles.

oblačić ključnih reči iz naslova radova, i sve to će biti zanimljivo za gledati.

All of that, I think, will be pretty interesting to see.

Odlična stvar je što to zapravo može i uraditi.

And the great thing is, it can actually do it.

Idemo, evo lepe krivulje zvona.

Here we go, a nice bell curve.

Vidite da je najčešće tri autora.

You see that three is kind of the most common.

It's going to then make this nice plot of the papers per year.

Napraviće lepu seriju radova po godini.

Something crazy is happening in 2023, though.

Nešto ludo se dešava ipak u 2023. godini.

Looks like we were on an exponential and it dropped off the cliff.

Eksponencijalno je rasla, a onda kao da se strmoglavila.

What could be going on there?

Šta se tu dešava? Usput, sve ovo je u Pajtonovom kodu, možete ispitati.

By the way, all this is Python code, you can inspect.

And then we'll see word cloud.

I evo nas onda kod oblačića ključnih reči, gde vidimo sve te divne pojmove iz radova.

So you can see all these wonderful things that appear in these titles.

But I'm pretty unhappy about this 2023 thing.

Ali prilično sam nezadovoljan ovim u 2023, zbog toga ova godina izgleda loše.

It makes this year look really bad.

Of course, the problem is that the year is not over.

Naravno, problem je što godina još nije gotova.

So I'm going to push back on the machine.

Pa ću da uzvratim mašini.

[Waitttt that's not fair!!!

[Čekaj, to nije fer!!! 2023. godina još nije gotova]

2023 isn't over.

[Koji procenat radova iz 2022. godine su objavljeni do 13. aprila?]

What percentage of papers in 2022 were even posted by April 13?]

So April 13 was the cut-off date I believe.

Verujem da je 13. april bio krajnji rok.

Can you use that to make a fair projection?

Možeš li to iskoristiti da napraviš precizniju projekciju?

So we'll see, this is the kind of ambitious one.

Vidićemo, ovo je prilično ambiciozno.

(Laughter)

(Smeh)

So you know,

Znate, opet, čini mi se da sam još toga želeo da izvučem iz mašine ovde.

again, I feel like there was more I wanted out of the machine here.

I really wanted it to notice this thing,

Stvarno sam želio da primeti ovo.

maybe it's a little bit of an overreach for it

Možda je malo za previše očekivati da magično zaključi da sam baš to želeo.

to have sort of, inferred magically that this is what I wanted.

But I inject my intent,

Ali izrazio sam svoju nameru, pružio dodatna uputstva,

I provide this additional piece of, you know, guidance.

And under the hood,

I pod haubom, VI i dalje piše kod, možete da preglete kako radi, veoma je moguće.

the AI is just writing code again, so if you want to inspect what it's doing,

it's very possible.

And now, it does the correct projection.

A sada, dala je ispravnu projekciju.

(Applause)

(Aplauz)

If you noticed, it even updates the title.

Ako ste primetili, čak je i ažurirala naslov.

I didn't ask for that, but it know what I want.

Nisam ni pitao, ali znala je da to želim.

Now we'll cut back to the slide again.

Vratimo se opet na slajd.

This slide shows a parable of how I think we ...

Ovaj slajd prikazuje parabolu o tome kako ja mislim da ćemo...

A vision of how we may end up using this technology in the future.

Viziju kako bi mogli da koristimo ovu tehnologiju u budućnosti.

A person brought his very sick dog to the vet,

Osoba je dovela svog veoma bolesnog psa veterinaru,

and the veterinarian made a bad call to say, “Let’s just wait and see.”

koji je doneo veoma lošu odluku i rekao: “Sačekajmo da vidimo šta će biti.”

And the dog would not be here today had he listened.

Psa ne bi danas bilo da ga je poslušao.

In the meanwhile, he provided the blood test,

U međuvremenu, dao je nalaz krvi i medicinski karton GTP-4,

like, the full medical records, to GPT-4,

which said, "I am not a vet, you need to talk to a professional,

koji je rekao: “Nisam veterinar, trebalo bi razgovarati sa profesionalcem,

here are some hypotheses."

ali evo nekih pretpostavki.”

He brought that information to a second vet

S tim informacijama je otišao po drugo mišljenje, i taj veterinar je spasio psa.

who used it to save the dog's life.

Now, these systems, they're not perfect.

Ovi sistemi nisu savršeni.

You cannot overly rely on them.

Ne možete se preterano oslanjati na njih.

But this story, I think, shows

Ali mislim da ova priča ilustruje

that a human with a medical professional

kako je čovek s medicinskim profesionalcem

and with ChatGPT as a brainstorming partner

i ChatGPT kao idejnim partnerom

was able to achieve an outcome that would not have happened otherwise.

postigao ishod koji se ne bi inače dogodio.

I think this is something we should all reflect on,

Mislim da je to nešto o čemu bi svi trebalo da razmislimo

think about as we consider how to integrate these systems

i razmotrimo prilikom integracije ovih sistema u naše živote.

into our world.

I jedna stvar u koju doboko verujem

And one thing I believe really deeply,

is that getting AI right is going to require participation from everyone.

je da će za usavršavanje VI biti potrebno učestvovanje svih nas.

And that's for deciding how we want it to slot in,

Kao i odlučivanje kako želimo da je integrišemo, postavimo joj pravila,

that's for setting the rules of the road,

for what an AI will and won't do.

šta VI sme i ne sme da radi.

And if there's one thing to take away from this talk,

Zaključak ovog govora je da ova tehnologija prosto izgleda drugačije.

it's that this technology just looks different.

Just different from anything people had anticipated.

Prosto je različita od svega što smo predviđali.

And so we all have to become literate.

Svi ćemo morati da se opismenimo.

And that's, honestly, one of the reasons we released ChatGPT.

Iskreno, to je jedan od razloga što smo objavili ChatGTP.

Together, I believe that we can achieve the OpenAI mission

Zajedno, mislim da možemo postići misiju OpenAI-a,

of ensuring that artificial general intelligence

a to je da opšta veštačka inteligencija koristi čitavom čovečanstvu.

benefits all of humanity.

Thank you.

Hvala vam.

(Applause)

(Aplauz)

(Applause ends)

(Kraj aplauza)

Chris Anderson: Greg.

Kris Anderson: Greg.

Wow.

Opa.

I mean ...

Hoću reći...

I suspect that within every mind out here

Slutim da se se u svačijoj glavi sada javlja osećaj zbunjenosti.

there's a feeling of reeling.

Like, I suspect that a very large number of people viewing this,

Slutim da veliki broj ljudi koji sada ovo gleda,

you look at that and you think, “Oh my goodness,

gleda i razmišlja: “Bože, gotovo sve o mom načinu rada, moraću ponovo da promislim.”

pretty much every single thing about the way I work, I need to rethink."

Like, there's just new possibilities there.

Prosto se otvaraju nove opcije, zar ne?

Am I right?

Ko sve misli da će morati ponovo da promisli o načinu na koji se nešto radi?

Who thinks that they're having to rethink the way that we do things?

Yeah, I mean, it's amazing,

Hoću reći, da, to je sve neverovatno, ali ujedno i zastrašujuće.

but it's also really scary.

So let's talk, Greg, let's talk.

Stoga, porazgovarajmo, Greg. GB: Naravno.

I mean, I guess my first question actually is just

KA: Moje prvo pitanje bi bilo, kako si dođavola ovo postigao?

how the hell have you done this?

(Laughter)

(Smeh)

OpenAI has a few hundred employees.

OpenAI ima par stotina zaposlenih.

Google has thousands of employees working on artificial intelligence.

Gugl ima hiljade zaposlenih koji rade na veštačkoj inteligenciji.

Why is it you who's come up with this technology

Kako si baš ti osmislio ovu tehnologiju koja je šokirala svet?

that shocked the world?

Greg Brockman: I mean, the truth is,

GB: Iskreno, svi gradimo na temeljima koje su postavili velikani, to je neupitno.

we're all building on shoulders of giants, right, there's no question.

Gledajući pomake u komputaciji, podacima, algoritmima, svi su na nivou industrije.

If you look at the compute progress,

the algorithmic progress, the data progress,

all of those are really industry-wide.

Mislim da smo sa OpenAI napravili mnoge ciljane izbore od samog početka.

But I think within OpenAI,

we made a lot of very deliberate choices from the early days.

Prvi je bio prosto da se suočimo sa realnošću kakva jeste.

And the first one was just to confront reality as it lays.

Zapitali smo se: “Šta će biti potrebno da se ovde naprave pomaci?”

And that we just thought really hard about like:

What is it going to take to make progress here?

We tried a lot of things that didn't work, so you only see the things that did.

Svašta smo probali, nije sve uspelo, ostale su samo stvari koje rade.

Najvažnije je bilo da se okupe timovi ljudi koji se razlikuju jedni od drugih,

And I think that the most important thing has been to get teams of people

who are very different from each other to work together harmoniously.

a koji mogu raditi složno.

CA: Can we have the water, by the way, just brought here?

KA: Možete li nam doneti vode?

Trebaće nam, biće ovo tema od koje se suše usta.

I think we're going to need it, it's a dry-mouth topic.

But isn't there something also just about the fact

Ima li to veze sa činjenicom da si video nešto u ovim jezičkim modelima iz čega bi

that you saw something in these language models

that meant that if you continue to invest in them and grow them,

ukoliko se u to uloži i razvije, moglo nešto nastati u jednom trenutku?

that something at some point might emerge?

GB: Yes.

GB: Da.

And I think that, I mean, honestly,

Iskreno, mislim da ova priča to sjajno ilustruje, zar ne?

I think the story there is pretty illustrative, right?

I think that high level, deep learning,

Mislim da duboko učenje, visokog nivoa, to je ono što smo uvek želeli.

like we always knew that was what we wanted to be,

Želeli smo laboratoriju za duboko učenje. Ali kako da to postignemo?

was a deep learning lab, and exactly how to do it?

U početku, mislim da nismo ni znali. Svašta smo probali.

I think that in the early days, we didn't know.

We tried a lot of things,

Neko je obučavao model kako da prepozna sledeće slovo u recenziji na Amazonu.

and one person was working on training a model

to predict the next character in Amazon reviews,

and he got a result where -- this is a syntactic process,

I došao je do rezultata gde, a govorimo o sintaktičkom procesu,

you expect, you know, the model will predict where the commas go,

očekujete da model predvidi gde će ići zarezi, imenice i glagoli.

where the nouns and verbs are.

But he actually got a state-of-the-art sentiment analysis classifier out of it.

Ali je zapravo došao do vrhunskog klasifikatora sentimentalne analize.

This model could tell you if a review was positive or negative.

Model vam je mogao reći da li je recenzija pozitivna ili negativna.

I mean, today we are just like, come on, anyone can do that.

Danas kažemo: “Ma daj, svako to može.” Ali tada je to bio prvi takav slučaj,

But this was the first time that you saw this emergence,

this sort of semantics that emerged from this underlying syntactic process.

takav tip semantike koja se pojavila u osnovi ovog sintaktičkog procesa.

And there we knew, you've got to scale this thing,

I tada smo znali, moramo to skalirati, videti dokle može da ide.

you've got to see where it goes.

KA: Mislim da će ovo rasvetliti misteriju koja muči sve koji ovo gledaju,

CA: So I think this helps explain

the riddle that baffles everyone looking at this,

jer ih često opisujemo kao mašine koje predviđaju.

because these things are described as prediction machines.

A opet, ono što vidimo da rade ...

And yet, what we're seeing out of them feels ...

it just feels impossible that that could come from a prediction machine.

Prosto se čini nemoguće da ovakvo nešto dolazi od mašine za predviđanje.

Just the stuff you showed us just now.

Sve ovo što si nam pokazao maločas.

And the key idea of emergence is that when you get more of a thing,

Sama ideja emergencije je da kada povećavate srazmeru ili učestalost nečega,

suddenly different things emerge.

prosto se pojave nove stvari. To se stalno deštava.

It happens all the time, ant colonies, single ants run around,

Npr, kolonije mrava. jedan mrav samo hoda unaokolo, ali ako ih je dovoljno,

when you bring enough of them together,

nastane kolonija koja pokazuje skroz drugačije, novonastalo ponašanje.

you get these ant colonies that show completely emergent, different behavior.

Or a city where a few houses together, it's just houses together.

Ili grad. Par kuća je samo par kuća zajedno.

Ali kako broj kuća raste, nastaju nove stvari,

But as you grow the number of houses,

things emerge, like suburbs and cultural centers and traffic jams.

poput predgrađa, kulturnih centara i gužvi u saobraćaju.

Give me one moment for you when you saw just something pop

Daj mi jedan trenutak u kom si video da nastaje nešto

that just blew your mind

što te je oduvalo, što prosto nisi predvideo.

that you just did not see coming.

GB: Yeah, well,

GB: Evo, probajte ovo u ChatGPT-u, sabirajte brojeve od 40 cifara.

so you can try this in ChatGPT, if you add 40-digit numbers --

CA: 40-digit?

KA: 40 cifara?

GB: 40-digit numbers, the model will do it,

GB: Brojeve od 40 cifara, i model će ih sabrati,

which means it's really learned an internal circuit for how to do it.

što znači da je formirao interno kolo kako da to radi.

And the really interesting thing is actually,

Ali zanimljiva stvar je, ako ga pitate da sabere brojeve od 40 i 35 cifara,

if you have it add like a 40-digit number plus a 35-digit number,

it'll often get it wrong.

često će pogrešno sabrati.

And so you can see that it's really learning the process,

Vidi se da uči proces, ali da ga nije još do kraja savladao, generalizovao, zar ne?

but it hasn't fully generalized, right?

It's like you can't memorize the 40-digit addition table,

Ne možete da naučite tablicu sabiranja za 40-cifrene brojeve

that's more atoms than there are in the universe.

to je više nego što atoma ima u univerzumu.

So it had to have learned something general,

Pa je morao da nauči nešto uopšteno, ali još to nije savladao do kraja.

but that it hasn't really fully yet learned that,

Oh, I can sort of generalize this to adding arbitrary numbers

U fazonu: “Sad mogu da generalizujem ovo, da sabiram brojeve proizvoljne dužine.”

of arbitrary lengths.

CA: So what's happened here

KA: Dakle, ovde se desilo to da ste mu dozvolili da poveća razmere

is that you've allowed it to scale up

and look at an incredible number of pieces of text.

i da sagleda neverovatne količine teksta.

And it is learning things

I sada uči stvari za koje niste verovali da će biti u stanju naučiti.

that you didn't know that it was going to be capable of learning.

GB Well, yeah, and it’s more nuanced, too.

GB: U suštini, ali se ide i u tančine.

So one science that we’re starting to really get good at

Jedna od nauka u kojima smo se poboljšali je predviđanje mogućih emergencija.

is predicting some of these emergent capabilities.

And to do that actually,

Da bismo to postigli, ne hvalimo dovoljno u ovom polju inženjerski kvalitet.

one of the things I think is very undersung in this field

is sort of engineering quality.

Like, we had to rebuild our entire stack.

Npr, morali smo da ponovo izgradimo naš čitav stek.

When you think about building a rocket,

Kada gradite raketu, tolerancije prema greškama moraju biti izuzetno male.

every tolerance has to be incredibly tiny.

Same is true in machine learning.

Isto važi i za mašinsko učenje.

You have to get every single piece of the stack engineered properly,

Svaki pojedinačni stek mora biti projektovan kako treba,

and then you can start doing these predictions.

i tek onda se može pristupi predviđanjima.

There are all these incredibly smooth scaling curves.

Tu je mnoštvo izuzetno glatkih skalirajućih krivulja.

They tell you something deeply fundamental about intelligence.

One vam saopštavaju nešto suštinsko o inteligenciji.

Pogledajte našu objavu na GPT-4 blogu i vidićete sve ove krivulje.

If you look at our GPT-4 blog post,

you can see all of these curves in there.

And now we're starting to be able to predict.

I sada već možemo da predviđamo.

Npr., možemo da predvidimo učinak na problemima u kodiranju.

So we were able to predict, for example, the performance on coding problems.

We basically look at some models

U suštini, gledamo neke modele koji su 10 000 ili 1 000 puta manji.

that are 10,000 times or 1,000 times smaller.

And so there's something about this that is actually smooth scaling,

Tako da tu ima nečega što je zapravo glatko skaliranje, iako je još u začetku.

even though it's still early days.

CA: So here is, one of the big fears then,

KA: I tako dolazimo do jednog od najvećih strahova koji proizilazi iz ovoga.

that arises from this.

If it’s fundamental to what’s happening here,

Ako je u osnovi ovoga što se dešava, to da kako povećavamo razmeru,

that as you scale up,

things emerge that

pojavljuju se stvari koje se mogu predvideti sa određenom dozom sigurnosti,

you can maybe predict in some level of confidence,

but it's capable of surprising you.

ali i dalje mogu da iznenade.

Why isn't there just a huge risk of something truly terrible emerging?

Zar ne postoji opasnost da se nešto stvarno strašno pojavi?

GB: Well, I think all of these are questions of degree

GB: Mislim da je sve to pitanje razmere i vremena.

and scale and timing.

I ono što ljudima takođe promiče je da je i integracija

And I think one thing people miss, too,

is sort of the integration with the world is also this incredibly emergent,

ove tehnologije takođe nova, a pri tome i izuzetno moćna pojava.

sort of, very powerful thing too.

To je jedan od razloga što mislimo da se ova tehnologija treba uvoditi postepeno.

And so that's one of the reasons that we think it's so important

to deploy incrementally.

Ono što mislim da sada vidimo, gledajući ovaj govor,

And so I think that what we kind of see right now, if you look at this talk,

je da se ja većinom bavim davanjem kvalitetne povratne informacije.

a lot of what I focus on is providing really high-quality feedback.

Danas se u to može ostvariti uvid, zar ne?

Today, the tasks that we do, you can inspect them, right?

Lako se može videti matematički problem

It's very easy to look at that math problem and be like, no, no, no,

i reći: “Ne, mašino, sedam je ispravan odgovor.”

machine, seven was the correct answer.

Ali čak i za rezimiranje knjige, to se teško nadgleda.

But even summarizing a book, like, that's a hard thing to supervise.

Like, how do you know if this book summary is any good?

Kako da znaš da ovaj siže ičemu valja? Moraš pročitati knjigu, a to niko ne želi.

You have to read the whole book.

No one wants to do that.

(Smeh)

(Laughter)

And so I think that the important thing will be that we take this step by step.

I zato mislim da će biti bitno da se stvari rade korak po korak.

And that we say, OK, as we move on to book summaries,

I da kažemo, u redu, kad pređemo na sažimanje knjiga,

we have to supervise this task properly.

moraćemo to propisno nadgledati.

We have to build up a track record with these machines

Moramo prvo izgraditi istorijat uspeha sa ovim mašinama

that they're able to actually carry out our intent.

da bi one uopšte bile u stanju da iznesu našu nameru.

And I think we're going to have to produce even better, more efficient,

Moraćmo pronaći još bolje, efikasnije i pouzdanije načine skaliranja

more reliable ways of scaling this,

sort of like making the machine be aligned with you.

toga da mašina bude usklađena sa vama.

CA: So we're going to hear later in this session,

KA: Čućemo kasnije govore gde kritičari kažu

there are critics who say that,

you know, there's no real understanding inside,

da nema istinskog razumevanja unutrašnjosti sistema, da će on uvek --

the system is going to always --

we're never going to know that it's not generating errors,

da nikada nećemo znati da ne pravi greške, da neće imati zdrav razum i tome slično.

that it doesn't have common sense and so forth.

Is it your belief, Greg, that it is true at any one moment,

Da li ti, Greg, veruješ da je to sada tačno,

but that the expansion of the scale and the human feedback

ali da će je povećanje razmere i povratne informacije od ljudi,

that you talked about is basically going to take it on that journey

o čemu si govorio, povesti u tom smeru

of actually getting to things like truth and wisdom and so forth,

i da će vremenom doći do istine, mudrosti,

i tome slično, sa većom sigurnošću? Kako možeš biti siguran u to?

with a high degree of confidence.

Can you be sure of that?

GB: Yeah, well, I think that the OpenAI, I mean, the short answer is yes,

GB: Pa, kratak odgovor je da. Mislim da se OpenAI kreće u tom smeru.

I believe that is where we're headed.

And I think that the OpenAI approach here has always been just like,

A naš pristup je uvek bio da pustimo realnost da nas ošine po faci, zar ne?

let reality hit you in the face, right?

It's like this field is the field of broken promises,

Ovo je oblast prekršenih obećanja,

gde stručnjaci govore da će se X desiti, a na način Y.

of all these experts saying X is going to happen, Y is how it works.

Ljudi su govorili da neuronske mreže neće raditi još 70 godina.

People have been saying neural nets aren't going to work for 70 years.

Još uvek nisu u pravu.

They haven't been right yet.

Možda će trebati vremenski period od 70 plus jednu godinu, ili nešto slično.

They might be right maybe 70 years plus one

or something like that is what you need.

Mislm da je naš pristup uvek bio da se granice ove tehnologije moraju gurati

But I think that our approach has always been,

you've got to push to the limits of this technology

da se vidi za šta je ona sposobna.

to really see it in action,

jer to nam govori kako i kada ćemo moći preći na novu paradigmu.

because that tells you then, oh, here's how we can move on to a new paradigm.

I mislim da ovde nismo još iscrpli sve.

And we just haven't exhausted the fruit here.

KA: Mislim da si zauzeo prilično kontroverzno stanovište,

CA: I mean, it's quite a controversial stance you've taken,

a to je da je pravi način da sve ovo staviš u javnost

that the right way to do this is to put it out there in public

i onda to sve iskoristiš,

and then harness all this, you know,

umesto da samo tvoj tim daje povratne informacije,

instead of just your team giving feedback,

the world is now giving feedback.

čitav svet ti daje povratne informacije.

But ...

Ali...

If, you know, bad things are going to emerge,

Ako se recimo pojave loše stvari, prosto će biti puštene u etar.

it is out there.

So, you know, the original story that I heard on OpenAI

Izvorna priča o OpenAI, kada ste osnovani kao neprofitna organizacija, je bila

when you were founded as a nonprofit,

well you were there as the great sort of check on the big companies

da ćete biti brana od velikih korporacija

doing their unknown, possibly evil thing with AI.

koje rade nepoznate, verovatno i loše stvari sa VI.

And you were going to build models that sort of, you know,

I da ćete graditi modele koji će ih nekako držati odgovornim

somehow held them accountable

and was capable of slowing the field down, if need be.

i biti u stanju da uspore oblast, ukoliko bude potrebno.

Or at least that's kind of what I heard.

Ili sam nešto slično tome čuo.

And yet, what's happened, arguably, is the opposite.

A zapravo se desilo upravo suprotno.

That your release of GPT, especially ChatGPT,

Puštanje GPT-a, a pogotovo ChatGPT-a

sent such shockwaves through the tech world

je toliko odjeknulo u svetu tehnologije

that now Google and Meta and so forth are all scrambling to catch up.

da se sada Gugl, Meta i drugi jagme da vas sustignu.

And some of their criticisms have been,

Neke od njihovih kritika su

you are forcing us to put this out here without proper guardrails or we die.

da ih silite da puste stvari bez propisnih mehanizama zaštite ili će propasti.

You know, how do you, like,

Kako opravdate ovo? Da je to urađeno na odgovoran način, a ne nesmotreno?

make the case that what you have done is responsible here and not reckless.

GB: Yeah, we think about these questions all the time.

GB: Da, ova pitanja su nam stalno na pameti.

Like, seriously all the time.

Ne, ozbiljno. Stalno.

And I don't think we're always going to get it right.

Ne mislim da ćemo uvek biti u pravu.

But one thing I think has been incredibly important,

Jedna stvar nam je uvek bila važna, i od samog početka smo razmišljali

from the very beginning, when we were thinking

kako da izgradimo veštačku opštu inteligenciju za dobrobit čovečanstva.

about how to build artificial general intelligence,

actually have it benefit all of humanity,

Mislim, kako da to uradimo?

like, how are you supposed to do that, right?

Neki plan koji se podrazumeva je da je izgradite u tajnosti, tu jako moćnu stvar,

And that default plan of being, well, you build in secret,

you get this super powerful thing,

i onda osmislite način sigurne upotrebe, pokrenete je i nadate se najboljem.

and then you figure out the safety of it and then you push “go,”

and you hope you got it right.

Ali ja ne znam kako da takav plan sprovedem. Neko drugi možda zna.

I don't know how to execute that plan.

Maybe someone else does.

Ali meni je to zastrašujuće, ne čini mi se ispravnim.

But for me, that was always terrifying, it didn't feel right.

Mislim da je ovaj alternativni pristup jedini drugi put koji ja vidim,

And so I think that this alternative approach

is the only other path that I see,

a to je da pustite da vas realnost ošine po licu.

which is that you do let reality hit you in the face.

I mislim da ljudima treba dati mogućnost davanja ulaznih podataka,

And I think you do give people time to give input.

da pre nego što ove mašine postanu savršene,

You do have, before these machines are perfect,

before they are super powerful, that you actually have the ability

postanu premoćne, da zapravo imamo uvid u to šta u praksi mogu da urade.

to see them in action.

And we've seen it from GPT-3, right?

A to smo videli kroz GPT-3, zar ne?

Kod GPT-3, plašili smo se

GPT-3, we really were afraid

that the number one thing people were going to do with it

da će glavna stvar za koju će se model koristiti biti

stvaranje dezinformacija i pokušaj uticanja na izbore.

was generate misinformation, try to tip elections.

Instead, the number one thing was generating Viagra spam.

Umesto toga, korišten je za stvaranje neželjene pošte za vijagru.

(Laughter)

(Smeh)

KA: Da, neželjena pošta za vijagru je loša, ali ima mnogo gorih stvari.

CA: So Viagra spam is bad, but there are things that are much worse.

Evo misaonog eksperimenta za tebe.

Here's a thought experiment for you.

Suppose you're sitting in a room,

Recimo da sediš u sobi. U sobi je sto, a na njemu kutija.

there's a box on the table.

You believe that in that box is something that,

Veruješ da je u toj kutiji nešto za šta

there's a very strong chance it's something absolutely glorious

postoji velika mogućnost da je nešto predivno

that's going to give beautiful gifts to your family and to everyone.

što će tebi, tvojoj porodici i svima dati divne darove.

But there's actually also a one percent thing in the small print there

Ali postoji i jedan posto šanse da negde sitnim slovima piše: “Pandora.”

that says: “Pandora.”

And there's a chance

I postoji šansa da se oslobode neslućena zla u svet.

that this actually could unleash unimaginable evils on the world.

Do you open that box?

Hoćeš li otvoriti tu kutiju?

GB: Pa, naravno da ne.

GB: Well, so, absolutely not.

I think you don't do it that way.

Ne mislim da tako treba uraditi.

And honestly, like, I'll tell you a story that I haven't actually told before,

Iskreno, ispričaću ti priču koju još nikom nisam ispričao,

which is that shortly after we started OpenAI,

a to je da kad smo pokrenuli OpenAI

I remember I was in Puerto Rico for an AI conference.

sećam se da sam bio u Portoriku na konferenciji o VI.

I'm sitting in the hotel room just looking out over this wonderful water,

Sedeo sam u hotelskoj sobi i gledao divnu vodu i ljude koji se zabavljaju.

all these people having a good time.

I čovek se zapita, ako biste imali izbor da otvorite tu vašu Pandorinu kutiju

And you think about it for a moment,

if you could choose for basically that Pandora’s box

to be five years away

za pet ili 500 godina, šta biste izabrali?

or 500 years away,

which would you pick, right?

On the one hand you're like, well, maybe for you personally,

S jedne strane, vama lično, možda je bolje za pet godina.

it's better to have it be five years away.

But if it gets to be 500 years away and people get more time to get it right,

Ali ako bi to bilo za 500 godina i ljudi bi imali vremena da to urade kako treba,

which do you pick?

šta onda izabrati?

And you know, I just really felt it in the moment.

I onda sam osetio u tom trenutku, naravno da bi izabrali da to bude za 500 godina.

I was like, of course you do the 500 years.

Brat mi je u to vreme bio u vojsci i rizikovao bi život na način

My brother was in the military at the time

and like, he puts his life on the line in a much more real way

mnogo stvarniji od bilo koga od nas kojih smo tipkali tada po računaru

than any of us typing things in computers

and developing this technology at the time.

i razvijali ovu tehnologiju.

And so, yeah, I'm really sold on the you've got to approach this right.

Tako da sam pristalica ideje da se ovome mora pristupiti kako treba.

But I don't think that's quite playing the field as it truly lies.

Ali ne mislim da je tu sve onako kako se na prvi pogled čini.

Like, if you look at the whole history of computing,

Ako se pogleda istorijat računarstva kao celine,

I really mean it when I say that this is an industry-wide

stvarno mislim kada kažem da je ovo što se sada dešava pomak na nivou industrije,

or even just almost like

ako ne čak i na nivou razvoja celokupne ljudske tehnologije.

a human-development- of-technology-wide shift.

And the more that you sort of, don't put together the pieces

Što duže ne povezujemo stvari koje su već u eteru,

that are there, right,

we're still making faster computers,

jer i dalje pravimo brže računare,

we're still improving the algorithms, all of these things, they are happening.

i dalje poboljšavamo algoritme, sve se to i dalje dešava,

And if you don't put them together, you get an overhang,

što duže to ne uvezujete, dolaziće do preklapanja,

što znači da ako neko, ili onog momenta kada neko uspe da to sve poveže,

which means that if someone does,

or the moment that someone does manage to connect to the circuit,

odjednom ćete imati jako moćnu stvar, a da niko nije imao vremena da se prilagodi.

then you suddenly have this very powerful thing,

no one's had any time to adjust,

Ko zna kakve ćete imati mere bezbednosti?

who knows what kind of safety precautions you get.

Moj zaključak je da, kada razmišljate o razvojima drugih tehnologija,

And so I think that one thing I take away

is like, even you think about development of other sort of technologies,

think about nuclear weapons,

npr., nuklearnog oružja,

people talk about being like a zero to one,

ljudi govore da je to bila promena iz nula u jedan

sort of, change in what humans could do.

u ljudskim mogućnostima.

But I actually think that if you look at capability,

Ali, ako pogledamo kroz njenu primenu, to je ipak bilo glatko kroz neki period.

it's been quite smooth over time.

And so the history, I think, of every technology we've developed

I istorijski gledano, svaka tehnologija koju smo razvili, razvijala se postepeno

has been, you've got to do it incrementally

and you've got to figure out how to manage it

i upravljali smo njome postepeno kako se povećavala njena upotreba.

for each moment that you're increasing it.

CA: So what I'm hearing is that you ...

KA: Dakle, ono što želiš da kažeš, da ovaj model koji želiš da imamo,

the model you want us to have

is that we have birthed this extraordinary child

da ga posmatramo kao jedno izuzetno novorođenče, koje možda ima supermoći,

that may have superpowers

koje mogu da povedu čovečanstvo do potpuno novog mesta.

that take humanity to a whole new place.

It is our collective responsibility to provide the guardrails

Da je naša kolektivna odgovornost da ovom detetu pružimo smernice, zaštitnu ogradu,

for this child

to collectively teach it to be wise and not to tear us all down.

da ga naučimo mudrosti i da nas ne uništi. Jesam li dobro shvatio tvoj model?

Is that basically the model?

GB: I think it's true.

GB: Mislim da si u pravu.

And I think it's also important to say this may shift, right?

Važno je napomenuti da se i to može promeniti.

We've got to take each step as we encounter it.

Svakoj fazi moramo pristupiti kako se s njom susrećemo.

And I think it's incredibly important today

Izuzetno je važno da danas svi postanemo pismeni u pogledu ove tehnologije,

that we all do get literate in this technology,

figure out how to provide the feedback,

shvatimo davanje povratnih informacija i odlučimo šta želimo od nje.

decide what we want from it.

And my hope is that that will continue to be the best path,

Nadam se da će ovo ostati najbolji put,

but it's so good we're honestly having this debate

i drago mi je da iskreno vodimo ovu raspravu

because we wouldn't otherwise if it weren't out there.

jer da model nije pušten, ne bismo je ni vodili.

CA: Greg Brockman, thank you so much for coming to TED and blowing our minds.

KA: Greg, hvala ti što si došao na TED i što si nas raspametio.

(Applause)

(Aplauz)